Skip to main content

2024 | Buch

Advances in Data-Driven Computing and Intelligent Systems

Selected Papers from ADCIS 2023, Volume 2

herausgegeben von: Swagatam Das, Snehanshu Saha, Carlos A. Coello Coello, Jagdish C. Bansal

Verlag: Springer Nature Singapore

Buchreihe : Lecture Notes in Networks and Systems

insite
SUCHEN

Über dieses Buch

This book is a collection of best-selected research papers presented at the International Conference on Advances in Data-driven Computing and Intelligent Systems (ADCIS 2023) held at BITS Pilani, K. K. Birla Goa Campus, Goa, India, during September 21–23, 2023. It includes state-of-the-art research work in the cutting-edge technologies in the field of data science and intelligent systems. The book presents data-driven computing; it is a new field of computational analysis which uses provided data to directly produce predictive outcomes. The book is useful for academicians, research scholars, and industry persons.

Inhaltsverzeichnis

Frontmatter
Deep Learning Models for Classification of Remotely Sensed Data of Sugarcane

The traditional machine learning algorithms are giving way to approaches for deep learning in computer vision, which refers to a computer's capacity to infer meaning from digital images and videos. Sugarcane categorization is important for agricultural management and monitoring. Traditional crop categorization methods based on manual inspection or restricted ground-based data gathering are time-consuming and frequently inaccurate. As a result, an automated and efficient strategy is suggested that requires the use of remote sensing data and the capabilities of deep learning algorithms. A dataset made from multispectral Sentinel imagery is used for the classification of sugarcane. This approach seeks to separate sugarcane-growing regions from other regions in Sentinel-2 images using VGG19, MobileNetV2, and CNN as feature extractors. These findings illustrate the feature extraction utilizing deep learning models with an SVM classifier for sugarcane. By considering variables such as distinct spectral bands, temporal fluctuations, and potential difficulties in separating sugarcane from other land cover types, the objective is to construct and check working of deep learning models for categorizing sugarcane locations using Sentinel-2 data. The sugarcane classification can further be used to find dense and sparse vegetation after the classification is done with deep learning models. The outcomes of this study will help to improve sugarcane categorization techniques and will help farmers, researchers, and agricultural stakeholders make better crop management, yield estimation, and resource optimization decisions in sugarcane farming.

Mansi Kambli, Bhakti Palkar
Detection and Analysis of Wormhole Attacks in the AODV Routing Protocol with IEEE 802.11p for the Internet of Vehicles

The Internet of Vehicles (IoV) is a network of vehicles that consists of vehicles, sensors, and technologies that allow communication between them. Its primary purpose is to enable vehicles to connect and share information via the Internet. The connectivity between all these entities is the most important. So, routing becomes the most crucial thing in IoV to establish connections. AODV is an essential protocol for efficiently connecting vehicles and other environments. In vehicular networks, the wormhole is a routing attack that severely threatens communication and data security. The objective of our research is to compare the performance analysis for metrics like Packet Delivery Ratio (PDR), End-to-End Delay (EED), and Throughput during wormhole attacks with those during normal operations. The study looks at how wormhole attacks affect the Ad hoc On-Demand Distance Vector (AODV) routing protocol. We have used NS 2.35 for simulation and getting results. We have also done statistical testing with one-way Analysis of Variance (ANOVA) to determine the significance of the difference. The study employs the IEEE 802.11p standard and aims to find evidence of the existence of wormhole attacks through statistical testing. The detection of wormhole attacks on IoV performance is explored through simulation results and statistical analysis.

Tanuj Meshram, Mou Dasgupta
A Systematic Review of NLP Applications in Clinical Healthcare: Advancement and Challenges

This systematic literature review examines the advancements and challenges of natural language processing applications in clinical healthcare. Authors provide an overview of NLP applications, including clinical text classification, named entity recognition, information extraction, clinical dialogue systems, and clinical decision support. These applications have improved clinical documentation, patient care, and research outcomes. Authors critically evaluate challenges such as data privacy, lack of standardized datasets, and domain-specific language models. Ethical considerations, interoperability, and potential biases in NLP algorithms are also discussed. This review highlights the current state of NLP in clinical healthcare, identifies areas for improvement, and suggests future research directions. By synthesizing existing literature, this paper contributes to a deeper understanding of NLP’s potential in transforming clinical practice.

Rachit Garg, Anshul Gupta
An Investigational Analysis of Automatic Speech Recognition on Deep Neural Networks and Gated Recurrent Unit Model

For thousands of years, communication has played a crucial role in human existence, development, and globalization. Speech recognition has several uses, including biometric analysis, education, security, health care, and smart cities. Many scientists have spent years studying how machine learning may be applied to speech processing, particularly voice recognition. But in recent years, researchers have concentrated on ways to apply deep learning to problems involving human speech. In this post, we discuss our work using deep neural networks like CRNN and GRU to recognize audio samples in spoken language. Seven different classes of audio samples (Walk & footsteps, Kids speaking, Filling with water, Bass drum, Scissors, Clock, and Cough) were employed in Free Sound Datasets. Mel-spectral coefficients, along with other spectral and intensity-related factors, are among the feature parameters utilized for recognition. White noise and a retuned voice were employed as data augmentation. An average recognition rate of accuracy 93.25% and WER—Word Error Rate—of 7.84% were obtained by the GRU model, according to the findings.

M. Soundarya, S. Anusuya
Matched Filter and Kirsch’s Template Based Approach for Retinal Vessel Segmentation

The analysis of retinal images plays an instrumental role in diagnosing Diabetic Retinopathy. This progressive disease can cause blindness which can be inhibited with earlier detection. A robust approach is presented in this paper for blood vessel segmentation by integrating a matched filter with Kirsch’s template with hysteresis thresholding. The proposed approach involved three steps: preprocessing, blood vessel extraction, and post-processing for vessel extraction. This approach achieved more accurate results than the original Kirsch’s template for specificity, sensitivity, and accuracy on the DRIVE dataset.

Sonali Dash, Kanwarpreet Kaur, Gaurav Bathla
Prediction of Abnormality in Kidney Function Using Classification Techniques and Fuzzy Systems

Kidney diseases are life threatening. Its development is prevented by early detection and vigorous management. It is important to discover such disorders at an early stage in order to extend a patient's lifespan and to classify the abnormalities in kidney function based on pathological data. The primary goal is to identify the stages of the kidney disease and check the performance for various classifiers of the model. In this paper, classification algorithms are used to find out the accuracy of the supervised data. Not all machine learning classifiers predict the accurate results because of imprecision. So, fuzzy expert system (FES) is used to deal with imprecise data. To predict the disease at an early stage and also to identify the stages of the disease, FES is used. FES has shown promising results in identifying the stages of the patients. The accuracy of the pathological data is found by using machine learning algorithms. In addition, the probability of the occurrence of the disease is found by combining various parameters and identified the stages of the patient’s disease.

Mynapati Lakshmi Prasudha, Sukhavasi Vidyullatha, Yeluri Divya
Implementation of Parallel Applications on the Hypercube Topology by Using Multistage Network

The recent computational systems include multicomputer formations. Multiple computers allow many tasks to be processed faster and concurrently and the possibility of implementing the same functions on various processors simultaneously. The problem is that companies are required to pay thousands of dollars to construct data centers, locations, servers, technicians, and hardware maintenance costs and over time need to update and upgrade. This paper simulates a virtual machine and uses the hypercube topology to implement a Multistage Network, a single computer is divided into eight computers as clients and eight computers as servers, the interconnection between servers is represented by mesh topology, and parallel processing is represented by multiplying two-dimensional matrices A and B. Cloud computing should virtualize the management of resources (such as memories, CPUs, storage) to users at reasonable costs. The goal of using this method is to construct a model that not only works on hypercube topology but also works on various topologies. The results showed that the outputs are computed through parallel virtual servers by using the threading technique. We produced a sample approaching a cloud computing system.

Qusay S. Alsaffar, Leila Ben Ayed
Integrating Artificial Intelligence for Adaptive Decision-Making in Complex System

The integration of AI techniques in systems engineering can revolutionize decision-making in complex systems. This research investigates AI's role in enhancing adaptive decision-making and addresses integration challenges. AI technologies, like machine learning and cognitive computing, handle large data volumes, identify patterns, and make accurate predictions, enabling decision-makers to gain valuable insights into system behavior, risks, and performance optimization. The research develops intelligent algorithms for real-time data analysis, pattern recognition, and anomaly detection, facilitating proactive decision-making. Intelligent decision support systems integrate AI technologies, providing real-time insights and recommendations to optimize performance and enhance system resilience. The research evaluates AI-based approaches through case studies, assessing their performance and effectiveness. It also addresses challenges such as data privacy, transparency, and system reliability, offering practical guidelines for successful AI integration. In conclusion, this research explores AI integration for adaptive decision-making in complex systems, advancing systems’ engineering and providing insights for practitioners and researchers implementing AI for effective decision-making in dynamic environments.

Ajay Verma, Nisha Singhal
Qualitative Research Reasoning on Dementia Forecast Using Machine Learning Techniques

The rise in mental health issues and the demand for high-quality medical care have prompted researchers to investigate how machine learning might be used to treat mental health issues. Dementia is a disease that causes loss of cognitive skills in a way that interferes with a person’s day-to-day activities. It causes a breakdown of brain function, comprehension, recognition, reasoning, and behavioral abilities to the point where a person experiences difficulties in day-to-day activities. Dementia gradually kills the brain cells and causes people to lose their reading and thinking capabilities. According to the Lancet report, the incidences of dementia cases in India are predicted to nearly triple by 2050. According to the survey, the number of cases is roughly predicted to quadruple to 153 million by 2050. This research presents the analysis and findings related to forecasting dementia using machine learning techniques. The study has been conducted using the Open Access Series of Imaging Studies (OASIS) dataset. This dataset has been explored by using various machine learning algorithms such as support vector machine, random forest, decision tree, logistic regression, AdaBoost, and XGBoost. The conclusion has been drawn regarding the evaluation metrics in accuracy. It has been found that XGBoost gave the best result with 93.02% accuracy. With XGBoost, it is simple to determine the ideal number of boosting iterations in a single run.

Tanvi Kapdi, Apurva Shah
Implementation of Vision Transformers on SPECT Heart Dataset: A Comparative Study

Medical imaging has gone through many changes and one of the changes could be observed since the introduction of transformers and deep learning models. Transformers have complicated architecture that can help in carrying out complex tasks on medical images or data. This research paper presents a comprehensive study on the implementation of the Vision Transformers (ViTs) on the Single-Photon Emission Computed Tomography (SPECT) heart dataset. The basic objective is to evaluate the accuracy of the classification of abnormal cardiac conditions from SPECT images using ViT. Further, this paper tries to explore different aspects of the architecture of transformers and their performance in comparison to the traditional machine learning models in analyzing the SPECT heart dataset. It helps to contribute to the growing community of research carried out in medical images and data using new technologies.

Poonam Verma, Vikas Tripathi, Bhaskar Pant
CSR U-Net: A Novel Approach for Enhanced Skin Cancer Lesion Image Segmentation

Early detection is very critical step in skin cancer diagnosis and treatment. This paper introduces a novel deep learning approach for skin cancer lesion image segmentation model, CSR U-Net: Channel–Spatial Regularized U-Net. The proposed model focuses on both channel attention and spatial attention with additional optimized regularization methods to prevent model overfitting. The paper discusses the implementation of U-Net, Attention U-Net, Residual U-Net models, and CSR U-Net and also compares the results. The segmentation task often has challenges due to variations in skin tones, quality of the image, variations in the lesion, noise, class imbalance, and boundary delineation. This research aims to create a better high performing model CSR U-Net that over comes the above-said challenges.

V. Chakkarapani, S. Poornapushpakala
Automatic Detection and Classification System for Mesothelioma Cancer Using Deep Learning Models with HPO

Mesothelioma is a deadly cancer, but its early detection is important to save the life of a human. Hence, the paper focuses on the development of a novel method to detect Mesothelioma cancer using deep learning techniques like Gated Recurrent Unit, Multilayer Perceptron, and Long Short-Term Memory along with GridSearchCV(a hyper-parameter optimization technique). To evaluate the method, an experiment has been conducted on the dataset of 324 records, where 228 represent healthy individuals and 96 depict Mesothelioma patients. After analyzing and studying its pattern, feature selection technique such as Standard Scaler is applied to remove extraneous attributes. Besides this, SMOTE technique has been also used to address class imbalance and balance the binary classes in the data. During model training, all the applied models have been trained as well as examined for the parameters like precision, accuracy, loss, F1-score, recall, and AUC-ROC. In addition to this, for enhancing the performance of MLP model, GridSearchCV has been incorporated to fine-tune the hyper-parameters. During experimentation, the results show that the MLP model incorporated with GridSearchCV optimizer achieves the highest testing accuracy of 98.97%, precision and AUC-ROC of 1.00, while as F1-score and recall of 0.98. These findings indicate that our proposed approach obtained through GridSearchCV demonstrates improved performance and serves as a reliable tool for early Mesothelioma detection.

Apeksha Koul, Rajesh K. Bawa, Yogesh Kumar
A Systematic Literature Survey on IoT in Health Care: Security and Privacy Threats

With the advancements in IoT technology, it is becoming a part of our life more than yesterday. IoT technology is revolutionizing the Healthcare sector by helping medical professionals in providing better care and remote monitoring to patients. Medical IoT devices generate very sensitive data from the patients, and security of this data is crucial for privacy of patients. Security solutions for Healthcare IoT systems need to be tailored specifically for these resource-limited IoT devices. This study surveys the various solutions given in the past few years for Healthcare IoT security. In this study, articles having solutions based on latest technologies such as blockchain, RFID are reviewed and analyzed to get better insights about the threats and their countermeasures’ results and features. In addition to this, the directions for future study are also mentioned. The aim of this study is to help everyone interested in the domain by providing a summary of latest schemes and generate interest among new scholars toward security concerns in Healthcare IoT.

Aryan Bakliwal, Deepak Panwar, G. L. Saini
Hybrid Deep Learning Framework for Glaucoma Detection Using Fundus Images

Glaucoma is a chronic eye condition that develops because intraocular pressure in the eye damages the visual nerve. One of the causes of blindness around the globe is due to it. Glaucoma does not initially cause vision loss, but if the condition worsens, it may leave a person permanently blind. Measurement of intraocular pressure, testing of the visual field, or inspection of the optical disc of fundus pictures are all methods used in the clinical setting to diagnose glaucoma. Early detection of glaucoma is crucial in reducing the risk of eye damage. VGG19, VGG19 + LSTM, Inceptionv3, and Inceptionv3 + LSTM are used to study the identification of glaucoma. ACRIMA is the dataset used, and it consists of 705 fundus images (396 glaucomatous images and 309 healthy images). The models are worked using data augmentation and K-fold cross-validation. The extracted features classify the input image as glaucomatous or healthy. The VGG19 + LSTM model performed the best out of all the models.

Royce Dcunha, Aaron Rodrigues, Cassandra Rodrigues, Kavita Sonawane
Sunflower Optimization with Elite Learning Strategy (SFO-ELS) for Antenna Selection in Massive MIMO Subarray Switching Architecture

Massive MIMO is a promising technology used by fifth generation of wireless technology to increase the channel capacity significantly. But the use of RF transceivers for every antenna at the base station increases hardware complexity and implementation cost of the system making it very challenging for deployment. This paper focuses on addressing the hardware complexity and cost challenges associated with Massive Multiple-Input Multiple-Output systems. To mitigate these challenges, an efficient antenna selection algorithm is essential for identifying a subset of antennas that contribute maximum to the channel capacity. By employing advanced antenna selection schemes, this study aims to identify the most effective approach for optimizing antenna selection in Massive MIMO technology. The Sunflower Optimization algorithm, combined with the elite learning strategy, offers a novel approach to antenna selection in Massive MIMO subarray switching architecture. It leverages the benefits of both the SFO algorithm and the elite learning strategy to improve the selection of antennas, thereby optimizing system performance. Sunflower Optimization with elite learning strategy has been proposed and evaluated the effectiveness of the simulated SFO algorithm by comparing it with traditional approaches. The result shows that the proposed method for antenna selection significantly improves upon other methods offering a more efficient and effective approach for enhanced channel capacity in a Massive MIMO system.

Snehal Gaikwad, P. Malathi
Machine Learning Models for Human Activity Recognition: A Comparative Study

With the advancements in machine learning, human activity recognition has found its applications in several emerging areas such as robotics healthcare, surveillance, smart environment etc. This paper aims to study and evaluate the performance of some popularly used machine learning algorithms in classifying human activities. We have selected K-NN, SVM and XGBoost methods in this study and the performance of the methods has been evaluated for 19 different activities which were performed by eight random persons. The required data was recorded using 5 MTx 3-DOF orientation trackers. The raw data was processed before feature extraction and then fed as input to the machine learning models. On performance comparison of these methods, it has been found that the SVM method when implemented with a polynomial kernel, outperforms the other state-of-the-art methods. It classified the different activities with an accuracy of 96.9%.

Anshul Sheoran, Ritu Boora, Manisha Jangra
A K-Means Variation Based on Careful Seeding and Constrained Silhouette Coefficients

K-Means is well-known clustering algorithm very often used for its simplicity and efficiency. Its properties have been thoroughly investigated. It is emerged that K-Means heavily depends on the seeding method used to initialize the cluster centroids and that, besides the seeding procedure, it mainly acts as a local refiner of the centroids and can easily become stuck around a local sub-optimal solution of the objective function cost. As a consequence, K-Means is often repeated many times, always starting with a different centroids’ configuration, to increase the likelihood of finding a clustering solution near the optimal one. In this paper, the Hartigan and Wong variation of K-Means (HWKM) is chosen because of its increased probability to ending up near the optimal solution. HWKM is then enhanced with the use of careful seeding methods and by an incremental technique which constrains the movement of points among clusters according to their Silhouette coefficients. The result is HWKM+ which, through a small number of restarts, is capable of generating a careful clustering solution with compact and well-separated clusters. The current implementation of HWKM+ rests on Java parallel streams. The paper describes the design and development of HWKM+ and demonstrates its abilities through a series of benchmark and real-world datasets.

Libero Nigro, Franco Cicirelli, Francesco Pupo
Satellite Image Analysis in Health Care—A Systematic Review

Rapid urbanization, population expansion, and escalating pollution levels have given rise to novel environmental concerns that demand the application of creative, analytical methodologies and diverse data sources. To ensure efficient urban ecological management, it is crucial to consider three levels of phenomena: the environmental system, the physical environment, and the regional surroundings. Utilizing remote sensing data is a viable avenue for enhancing the global environment due to its convenient accessibility and ability to measure crucial physical features regularly. The primary objective of this research study is to ascertain and evaluate suitable methods for analyzing satellite imagery in the healthcare context. The study’s aims encompass examining the diverse array of tools accessible in the market, scrutinizing the characteristics of each device, exploring research papers that have employed these tools, and classifying the tools into open-source and commercial classifications. This study emphasizes the parameters associated with feature extraction and picture enhancement in satellite imagery, as these factors hold significant importance in the analysis of images.

Bhushan Pawar, Vijay Prakash, Lalit Garg, Charles Galdies, Sandra Buttigieg, Neville Calleja
Achieving Sustainability in Supply Chain During Disruption Times: Role of Industry 4.0

The demand to digitalize the automotive sector, which entails linking manufacturers to a larger supply chain, is increasing. In the automobile industry, there is a higher requirement for sustainable growth due to rising supply disruption and frequent technology changes. Industry 4.0 can speed up manufacturing, increase customizability, and cut down on setup and lead times. It may result in innovation. The study focuses on establishing link between Industry 4.0 technologies and green supply chain practices, which will help in achieving sustainability during disruption times. It follows a qualitative survey approach to identify the prominent Industry 4.0 technologies and green supply chain practices using a fuzzy set analytical hierarchy process. The other section uses interpretive structural modelling with a multi-level hierarchical structure to study the cause-and-effect relationship between the final selected Industry 4.0 technologies and GSC practices. The study identifies a strong linkage between Industry 4.0 technologies and green supply chain practices to achieve overall sustainability in the supply chain. The future automotive supply chain should focus on driving Industry 4.0 technologies for effective implementation of green supply chain practices. Also in the Indian automotive sector, government regulation and policies and top management commitment are two key factors for driving sustainability in the Indian automotive supply chain.

Namit Shrivastava, Manoj K. Srivastava
Emotionally Engaged Neurosymbolic AI for Usable Password Generation

Password-based authentication remains essential despite the advent of Multi-factor Authentication (MFA). A significant challenge is encouraging users to create strong, memorable passwords, as weak or reused passwords pose considerable security risks. This research introduces the Emotionally Engaged Neurosymbolic AI (EENAI) system, a novel approach for generating usable passwords. It combines neurosymbolic AI and emotional engagement, utilizing valence and arousal in emotionally engaged scenarios. Neurosymbolic AI combines neural network learning with classical AI's symbolic reasoning, ideal for generating context-aware passwords. By integrating valence and arousal, EENAI generates secure, memorable passwords. This paper details EENAI principles, experimental procedures, and observations. Results suggest that EENAI-generated passwords balance security, memorability, and usability, potentially revolutionizing password creation practices. The passwords are further evaluated using a standard password strength estimation tool, yielding promising results. The paper concludes with an EENAI impact assessment and future work recommendations.

Sumitra Biswal
Exploring the Deep Learning Techniques in Plant Disease Detection: A Review of Recent Advances

In agriculture, protecting crop yield is one of the most critical aspects of avoiding crop waste and ensuring food security around the world. One of the most critical aspects of preserving yield is protecting it from pests and plant diseases. With the advancement in the field of Artificial Intelligence (AI), it has been applied to different domains, and one such field is agriculture, where we can incorporate AI. Deep learning (DL), which is a subset of Artificial Intelligence, has gained lots of attention toward plant disease detection in the present day because of its better accuracy and performance in comparison with other techniques like machine learning (ML), etc. In this paper, we provide a comprehensive review of the current research work by utilizing deep learning for plant disease detection. We study the different models and architectures proposed by different authors and try to identify the pros and cons of the proposed methodology. We also discuss the various datasets that have been used in research work for detecting plant diseases. Finally, we describe the possible challenges in implementing deep learning models and discuss the future roadmap that can be followed by trying to identify the research gaps.

Saurabh Singh, Rahul Katarya
A Robust Driver Distraction Estimation Technique for ADAS Applications

Road accidents account for significant economic and personal costs, and the cognitive state of the driver is one of its major causes. We propose a novel approach for a driver monitoring system (DMS) to detect the cognitive state of the driver in real time using object detection. For this, we have used data from 19 different drivers in diverse traffic conditions in Bengaluru with over 7 h of driving time. We have chosen the YOLOv5 algorithm for our classification model with three categories, namely focused, sleepy (eyes closed or yawning), and distracted (looking away from the road). A magnetometer integrated with this model effectively categorizes distracted head turns while discerning them from deliberate and desirable head movements; this improves the robustness of classification. Velocity is computed using an onboard GPS unit, and the readings are used to determine if the vehicle is stationary in which case the detections are ignored. Experiments conducted on 19 people showed that our system has an average accuracy of 96.90% for this three-class classification model.

Sriman Sathish, S. Ashwin, S. Manish, Nishanth S. Shukapuri, Mayur S. Gowda, Viswanath Talasila
A Hybrid Approach for Depression Detection Using Word Embedding, Naive Bayes and Bi-LSTM Models

Depression is a serious illness that negatively affects health and well-being. A large population suffers from depression and they do not want to talk about the mental illness. The stigma associated with mental illness may discourage people from getting treatment thus leading to serious issues, such as social isolation, discrimination and self-harm. The high use of social media enables people to express their feeling and thoughts easily. The objective of this research is the diagnosis of depression in a person from his/her social media behaviour. The novel approach of the proposed model is to ensemble the Gaussian Naive Bayes classifier and Bi-LSTM to find contextual semantics of the text using Part-of-Speech (POS) tagging and Word Embedding. The experimental result shows the proposed model outperforms the state-of–the-art method and shows an accuracy of 83% on the benchmark dataset.

Jyoti Singh, Ishan Mangotra, Minni Jain, Amita Jain
A Deep Learning Approach to Computer-Aided Screening and Early Diagnosis of Middle Ear Disease

This article introduces a deep learning approach to computer-aided screening and early diagnosis of middle ear diseases such as earwax, otitis externa, tympanosclerosis, and ear ventilation tubes. The timely detection of middle ear conditions is crucial for effective treatment and prevention of complications. The proposed system utilizes a deep neural network trained on a large dataset of middle ear images obtained through advanced diagnostic imaging techniques. The system automatically analyzes these images by leveraging deep learning to provide accurate and efficient screening and diagnostic support. The proposed system aims to assist healthcare professionals in accurate and efficient early diagnosis and screening of critical conditions, leading to improved patient outcomes and optimized treatment plans. The proposed model presents a deep learning 2D-CNN model for binary and multi-class classification of ear diseases in medical healthcare. The results demonstrate its effectiveness and superiority compared with the traditional machine learning approaches.

Ankit Kumar Singh, Ajay Singh Raghuvanshi, Anmol Gupta, Harsh Dewangan
Pay-by-Palm: A Contactless Payment System

Current payment systems, including cash, credit cards, and UPI can be inconvenient for users, prompting the need for a more robust and user-friendly payment system. Biometric authentication methods like palm prints can enhance security and the user experience, but there is a lack of a reliable system that integrates palm print recognition with e-wallets to facilitate payments at participating merchants. Existing payment systems fail to provide a secure and convenient way to pay using palm prints, with challenges regarding the accuracy, reliability, and privacy of palm print recognition technology. By integrating palm print recognition technology with e-wallets, this work aims to meet the growing demand for a more advanced payment system that enhances the user experience while providing a secure way to make payments.

Sridevi Saralaya, Pravin Kumar, Mohammed Shehzad, Mohammed Nihal, Pragnya Nagure
Steganalysis of Reversible Digital Watermarking Algorithm Based on LWT and SVD

The processes of Digital Rights Management are used to limit access to proprietary and copyrighted material. Original creator of any digital data needs protection against his rights. Digital Rights Management involves two main concepts, watermarking and cryptography. We are working with reversible digital watermarking to keep cover and hidden data safe after extracting hidden message. In this work, digital image data is used as a mask, and using MD5 (Message Digest) algorithm, we have generated a digest of data. The specific digital data ID is created. In an embedding algorithm, this watermark is then integrated with digital data. Here, we worked on a reversible embedding algorithm that enabled us to use the cover even after the secret message had been removed. We have considered the different kinds of digital watermarking and the form of cover data in this report. Due to imperceptibility features, a watermarked data may be transmitted over the open network. It will therefore draw multiple attacks and must be robust. We have used LWT-QR-MD5 techniques that are a lossless approach. The degree of robustness is improved when a consistent procedure is used to build the digest of the watermark. The results are showing a remarkable improvement over LWT.

Geeta Sharma, Vinay Kumar
Analyzing the Effectiveness of Image Augmentation for Soybean Crop and Broadleaf Weed Classification

Data is the key for every artificial intelligence (AI)-based application irrespective of the type of data (numerical, categorical, image) being used. Quality and the depth of information in the data determine the performance of the AI model. Before the data is given as input to the classifier, it must be cleaned using pre-processing techniques. The data must also be sufficient enough to produce satisfactory results. The data considered for this work is images and thus emphasizes image augmentation techniques. This paper focuses on analyzing the best image augmentation techniques for deep learning classifiers. It is essential to analyze the effective data augmentation technique for a particular dataset. In this work, 2382 observations (images) from a crop-weed dataset are used to build the classifier. To expand this dataset, 11 image augmentation methods are applied to the training images. Six out of 11 methods show a high level of effectiveness and are chosen for further process. The outcome of every augmentation method is depicted for an in-depth understanding of augmentation techniques. Sixteen convolutional neural network (CNN)-based pre-trained models are built for evaluating the results. However, MobileNet outperformed other models by resulting in an overall accuracy of 99.58% and F1score of 1.0. Moreover, the performance of the model is evaluated using 24 metrics, and the formulas used for calculation are also tabulated in detail. Tables and graphs are represented for understanding the outcome precisely. Future works in image processing with deep learning are also discussed before concluding.

Michael Justina, M. Thenmozhi
Iterative Thresholding-Based Shadow Detection Approach for UAV Images

Shadow detection is a critical task in computer vision and image processing that aims to identify shadow regions in an image. Accurate detection of the shadow is essential for various applications, such as object recognition, scene understanding, and image segmentation. The detection of shadow is difficult due to their complex and dynamic nature, as they can vary in shape, size, and intensity depending on the location of the illumination source, weather conditions, and the characteristics of the scene. In this study, a new shadow detection method has been proposed that automatically calculates the threshold value using an iterative thresholding scheme and detects shadow. The performance of the developed method is tested on four publicly available UAV image datasets related to two study areas namely urban and mining areas. The comparison of the proposed method with several state-of-the-art methods demonstrates that the proposed method performs well in both qualitative and quantitative evaluations, with good overall accuracy in all images.

Deeksha, Toshanlal Meenpal
A Subtle Design of Prediction Models Using Machine Learning Algorithms for Advocating Selection and Forecasting Sales of Garments: A Case Study

In this article, the predictive analysis is conducted for a garment retail dataset that contains the attributes of the dresses and sales information. Precisely, Random Forest (RF), Linear Regression (LR), Support Vector Machine (SVM), and Decision Tree (DT) algorithms are used for classification. That is, advising whether the dresses should be kept in store or not by automating the process of the recommendation. Moreover, two variants of the datasets are given as input to the said algorithms apart from the raw dataset. One variant is obtained through feature selection and another uses the concept of dummy variable since the majority of the features are categorical. In addition, the demand for sales is estimated over a period. Auto-Regressive Integrated Moving Average (ARIMA) is applied in particular to achieve the forecasting of the sales. The dataset contains fourteen features of dresses and sales data of alternative days over a month. The experiments on the case study show that RF algorithm is good at the classification although it is marginally better than LR. Also, the sales forecasting is producing results in an acceptable range as per the relevant performance metrics. Overall, the proposed methodology of this paper helps in the decision-making of fashion retail.

Dillip Rout, Bholanath Roy, Prasanna Kapse
BGKnow-Medical Chatbot: A Hybrid Approach Based on Knowledge Graph and GPT-2

Accurate and timely diagnosis is critical in ensuring patients receive care and treatment for their medical conditions. Traditional symptom checkers often lack accuracy and efficiency in diagnosing diseases, as they typically rely on preprogrammed decision trees or rule-based algorithms that may not account for the complexity and variability of symptoms. By using natural language processing (NLP) and machine learning techniques, such as knowledge graphs and Bio-Bidirectional Encoder Representations From Transformers (BioBERT), chatbots can provide a more accurate and personalized approach to disease diagnosis. This paper introduces a hybrid chatbot framework called “BGKnow.” BGKnow represents the combination of a knowledge graph and Generative Pre-trained Transformer 2 (GPT-2) model that uses BioBERT embeddings for effective diagnosis based on symptoms entered by users. The proposed system shows promising potential for assisting healthcare professionals in accurately and efficiently addressing medical inquiries.

Disha Sunil Nikam, D. Nisha Murthy, Sreeramya Dharani Pragada, H. R. Mamatha
Video Integrity Checking Using X25519 and Nested HMAC with BLAKE2b

Latest developments in video editing or manipulation tools have facilitated effortless alteration of video content without any discernible traces left behind. As a result, it is imperative to subject video data to an integrity verification process before utilizing it as evidence. This paper presents a novel, lightweight approach for verifying the integrity of video data. The proposed method uses Hash-based Message Authentication Code (HMAC) and Elliptic Curve Diffie-Hellman Key Exchange utilizing Curve 25519 (X25519) with BLAKE2b. Nodes store video verification codes that are generated for video clips of a specific predetermined size. To enhance the security level, each node stores the nested HMAC value of the prior node. The integrity check involves the regeneration and comparison of nested HMAC of the node. The proposed method’s experimental results demonstrate better performance in terms of both speed and security when compared to state-of-the-art methods. With minimal additional storage requirements, our method can identify any type of forgery on any video file, at any given time, by an authorized individual. Security analysis indicates that the method can withstand a range of attacks, such as timing attacks, key substitution attacks, side channel attacks, and brute force attacks.

Linju Lawrence, R. Shreelekshmi
Cloud-Based Skin Cancer Classification: Training and Deploying a Model on AWS

Skin cancer has become one of the most dangerous and most common types of cancer in recent years. Skin cancers come in a variety of types, and identifying the type is crucial for treating the condition when it is still treatable. The dermatologist must also distinguish between skin conditions that affect the tissues on the top layer of the skin and cells of the skin cancer that develop in the epidermal layer of the skin. The current methods for identifying or categorizing skin cancer take a long time and can be painful for the patient due to potential side effects. There is extensive research going on in this area but the unavailability of the balanced datasets and small size of the datasets have become a hindrance. There are not many products like web applications which use deep learning models to identify and categorize the type of skin cancer. We have used the ConvNeXt Tiny deep learning model which is pre-trained on ImageNet. Once we had obtained good accuracy, we had used that model and created a web application which is hosted in the AWS Cloud. The dataset we used for our work is ISIC2018. It consists of seven classes of dermoscopic images of skin lesions which are of high resolution. It is an imbalanced data that has images collected from various clinical sites.

Challa Koti Reddy, Chava Pavan Kumar, A. R. P. S. Gowtham, Rajkumar Maharaju, Rama Valupadasu
An Automotive ECU-Based Forward Collision Prevention System

This study presents a software model that identifies vehicles in front of a test vehicle, measures distances, and classifies them as safe, slow speed, and brake. The classification determines which signal should be transmitted to the control unit. A specially tailored dataset of 2162 images from Bangladesh’s roadside is used for the software model, which uses transfer learning to identify frontal objects, estimate distances, and classify distances according to control unit signals. Furthermore, two microcontrollers are used for hardware systems, utilizing an ultrasonic sensor to calculate distances, identify frontal objects, and show the expected outputs. The AURIX TC375 microcontroller board-control unit receives signals and triggers the appropriate output. This system can serve as a foundation for autonomous vehicle safety research.

Fariya Islam, Tajruba Tahsin Nileema, Fazle Rabbi Abir, Tasmia Tahmida Jidney, Kazi A. Kalpoma
Credit Card Fraud Detection by Using Ensemble Method of Machine Learning

Online transactions have become an essential aspect of life as universe becomes more technological and every industry leverages the web to grow enterprises. Online transactions have been increasing steadily, and this trend is expected to continue. Credit cards are a popular form of internet transaction, but with their widespread use comes a significant drawback: credit card fraud. Since banks are unable to screen every transaction, machine learning is essential to identifying credit card fraud. In our research, we used Kaggle to gather a dataset of 2,844,808 credit card transactions from a European Bank Dataset. There are 492 fraudulent transactions in it; to balance the dataset, we proposed hybrid resampling method; and for the detection of credit card fraud, Random Forest Algorithm is used. The assessment of the model is assessed based on accuracy, precision, recall, and F1-score. Our model shown fairly good results of 97.66, 98.85, 95.94, 97.37% for accuracy, precision, recall, and F1-score, respectively.

Nihar Ranjan, G. S. Mate, A. J. Jadhav, D. H. Patil, A. N. Banubakode
APiCroDD: Automated Pipeline for Crop Disease Detection

This research paper proposes APiCroDD: automated pipeline for crop disease detection, an automated framework for early detection of plant diseases using multispectral imagery from drones. Current frameworks for disease detection are labor and time-consuming. They do not leverage the richness of multispectral imagery for feature extraction and perform vanilla manipulation of agriculture indices. Our framework comprises two stages: data acquisition and disease identification. We find that the use of multispectral imagery in the proposed framework provides several advantages over traditional RGB imagery, including better spectral resolution and increased sensitivity to subtle changes in plant health. The multispectral data enables the identification of specific spectral bands associated with diseased regions of the plant, improving the accuracy of disease detection. The proposed framework utilizes a combination of CNNs and segmentation techniques to identify the plant and its disease. Experimental results demonstrate that the proposed framework using EfficientNet is highly effective in identifying a range of plant diseases achieving state-of-the-art performance on manually collected dataset and validated on the PlantVillage dataset.

Pawan K. Ajmera, Sanchit M. Kabra, Anish Mall, Ankur Lhila, Aaryan Agarwal
TimeGAN for Data-Driven AI in High-Dimensional Industrial Data

The availability of historical process data in predictive maintenance is often insufficient to train complex machine learning models. To address this issue, techniques for data augmentation and synthesis have been developed, including the use of Generative Adversarial Networks (GANs). In this paper, the authors apply the GAN-based approach to synthesize simulated time-series data. Experiments are carried out to find a trade-off between the amount of labeled data needed and the accuracy of the synthetic data for downstream tasks. The authors find that using 40% of the original data for training the GAN results in synthetic data containing the same information for downstream tasks as the original data, leading to an estimated speedup of 60% in the initial computing time. The results of the evaluation for the authors’ own FEM simulation data, as well as for the Tennessee-Eastman benchmark dataset, are presented, demonstrating the potential of GANs in reducing time and energy in process development, while additionally interpolating a fixed parameter grid that is subsequently used for simulation purposes. This work demonstrates the feasibility of using GANs for the generation of high-dimensional time-series data in industrial applications as a supplementary method to classical FEM simulations.

Felix Neubürger, Yasser Saeid, Thomas Kopinski
A Deep Learning Framework for Assamese Toxic Comment Detection: Leveraging LSTM and BiLSTM Models with Attention Mechanism

As social media platforms grow in popularity, this research piece discusses the significance of creating a secure and positive online environment. The major goal is to protect users by detecting objectionable language in Assamese social media comments. The ultimate goal is to create a very effective mechanism for detecting toxic comments in Assamese, supporting a safe online environment. To address the lack of available datasets, a well-curated dataset was manually assembled for the experiment. Deep learning models such as LSTM and bidirectional LSTM (BiLSTM) were used to capture the contextual intricacies of user-generated comments. Notably, the BiLSTM model beats the LSTM model by including an attention mechanism, attaining a promising accuracy rate of 86.9% in successfully identifying toxic comments. Using the capabilities of the LSTM and BiLSTM models, a more robust and efficient approach for recognizing toxic phrases in Assamese is developed, aligned with the goal of building a secure, respectful, and toxic-free online environment.

Mandira Neog, Nomi Baruah
Security in VANETs with Insider Attack Resistance and Signature Aggregation

Vehicular Ad-hoc Networks (VANETs) are a type of network in which vehicles communicate with one another and exchange information so as to provide quality-of-life improvements to the vehicle users as well as the people belonging to the area. In VANET, vehicles share information such as the current status of the vehicle, the status of the traffic or the status of the road conditions to the other vehicles. All this information requires the network to be highly secure. Therefore various schemes have been proposed to secure the network. However, they suffer when a receiver has to verify multiple incoming message signatures. To reduce the verifying time and size of the signature The proposed scheme provides an efficient pairing-free aggregate signature, it will verify multiple messages by combining multiple signatures into a single signature called aggregate signature, and it is also resistant to insider attacks without using the tamper-proof device.

Vijaya Lode, Kekhelo Lasushe, Anil Pinapati
Comparative Study of LevelDB and BadgerDB Databases on the Basis of Features and Read/Write Operations

Due to advent of huge complex datasets, key-value databases have achieved great demand for accessing data quickly and efficiently in comparison with relational databases. There are several key-value databases available, e.g., Level DB, Badger DB, MongoDB, etc. The data stores as a key—value pair that is so these databases perform all the operations on the dataset rapidly. In this paper the author’s purpose is to focus on two most famous key-value databases: Badger DB and Level DB and analyze the performance of both the databases. For this analysis the author created 10 different datasets for every read/write operation. This analysis study is based on the results carried out by instantiate, read, and write operations on these databases and therefore resulting how level DB is more efficient than Badger DB during read and write operations.

Pragya Vaishnav, Linesh Raja, Aniket Bhange
An Improved Snow Ablation Optimizer for Stabilizing the Artificial Neural Network

Artificial neural networks give more promising and accurate results than other methods for prediction, classification, and segmentation engineering problems. The accuracy of the artificial neural network is affected by the training algorithm used. Gradient-based optimization algorithms are traditional methods to train artificial neural networks. They find an accurate solution to the problem. However, they are sensitive to initial values. It makes them unstable for finding better accuracy results. Moreover, training time becomes higher. To overcome these problems, we proposed an improved snow ablation optimizer (ISAO) algorithm and used it to find the pre-trained weights and biases for initializing the artificial neural network’s weights and biases. Its performance was tested on the MNIST data set and compared with SGDM-BP, SAO, and GOA algorithms. The improved ISAO algorithm achieved better results than compared algorithms regarding cross-entropy, testing, and training accuracy.

Pedda Nagyalla Maddaiah, Pournami Pulinthanathu Narayanan
Backmatter
Metadaten
Titel
Advances in Data-Driven Computing and Intelligent Systems
herausgegeben von
Swagatam Das
Snehanshu Saha
Carlos A. Coello Coello
Jagdish C. Bansal
Copyright-Jahr
2024
Verlag
Springer Nature Singapore
Electronic ISBN
978-981-9995-21-9
Print ISBN
978-981-9995-20-2
DOI
https://doi.org/10.1007/978-981-99-9521-9