Proceedings of the 7th International Conference on Communications and Cyber Physical Engineering

Table of Contents

1. Unmasking Cyberbullying Through NLP-Driven Detection in Contemporary Social Media

   Sai Kowshik Goud, Bandari Dinesh Yadav, Raparthi Srennik Goud, Y. Mrudula, Raj Kumar Chanda, Pavan Kumar Pagadala

   Abstract

   Nowadays, cyberbullying is a really risky circumstance. As a result of experiencing a lot to handle, it leads young people to commit suicide. We have developed an algorithm that, upon receiving the messages or comments as input, reads them and analyzes whether they are abusive or not. Contemporary social networking sites are so simple to use that there have been many severe incidences of cyberbullying. Social media is extremely dangerous since people who aren’t even majors use it. Cyberbullying is defined as any offense against a person’s race, gender, culture, or religion. To find bullying-related keywords in the corpus, we created an algorithm using Natural Language Processing (NLP) and machine learning techniques. These techniques together work tremendously and help us build a project based on cyberbullying detection.

2. Arbitrary Leaf Image Focus Generation and Enhancement with CycleGAN and Super Resolution

   V. Grishma Neha Chowdary, Bitta Hari Charan, K. Sree Suryakanth, Raj Kumar Chanda, Pavan Kumar Pagadala, P. Lalitha Surya Kumari

   Abstract

   In traditional photography's limitations, where focus is fixed now of capture, this research uses CycleGAN to enable post-capture focus adjustments, enhancing both the correction of focus errors and the creative modification of images. This study explores the innovative use of Cycle-Consistent Generative Adversarial Networks (CycleGAN) for generating arbitrary focus images, specifically concentrating on leaf imagery. The research employs a two-part neural network architecture inherent in Generative Adversarial Networks (GANs), comprising a generator and a discriminator. These networks are trained concurrently, where the generator aims to produce increasingly realistic images, and the discriminator evaluates their authenticity. CycleGAN’s unique feature of cycle consistency is critical for image-to-image translation tasks, particularly useful in scenarios where paired examples for training are not available, such as in focus manipulation. In this study, CycleGAN is utilized to transform the focus of leaf images by generating dense focal stacks from sparse ones, demonstrating the model’s ability to learn domain transformations from near-focus to far-focus images and vice versa. The process involves subjective evaluations using the naked eye and objective evaluations using metrics like PSNR (Peak Signal-to-Noise Ratio). However, challenges such as significant color tone changes and deterioration in image quality were observed, indicating the model's tendency to learn using color features rather than focus. To enhance image quality, the study incorporates super-resolution methods like VDSR and EDSR, which use residual learning to deepen the model for more efficient training and to mitigate the risk of vanishing gradients. The results demonstrate a substantial improvement in CycleGAN’s accuracy, reaching 97.8%, with PSNR values indicating the model's effectiveness in generating high-quality images.

3. Empowering E-Commerce: Personalized Product Recommendations Through ML-Based Algorithms

   Peruri Anusha, Gaddala Greeshma Devi, Pavan Kumar Pagadala, Chanda Raj Kumar, Chiranjeevi Nuthalapati, Vinod Kumar Dharavath

   Abstract

   Personalized product recommendations have become a cornerstone in elevating user satisfaction and boosting e-commerce sales. The advent of Machine Learning Techniques has empowered the creation of personalized product selection models. This research embarks on an exploration of the development and deployment of ML-based algorithms tailored to generate personalized product recommendations. The primary objective is to enhance user engagement and bolster conversion rates. This study harnesses use]. Recommendations-encompassing elements like browsing history, purchase records, and demographic information to construct precise and effective recommendation models. Recommendation systems heavily depend on user data, aggregating past interactions, behaviors, and preferences, which are continuously gathered and archived for comprehensive analysis. Recommendation systems come in various forms. In contrast, content-based filtering considers item attributes and user preferences.

4. A Study on Drastic Climate Changes Leading to Global Warming

   Karnati Vamshi, M. T. V. Soumith, Vaibhav Venkateswaran, Kanaparthy Tribhuvan, Pavan Kumar Pagadala, Chanda Raj Kumar

   Abstract

   This project employs Exploratory Data Analysis (EDA) techniques to investigate climate feedback mechanisms, both positive and negative, and their influence on the carbon cycle. Utilizing Naive Forecasting, ARIMA model, and K-Means model, the study aims to comprehend the intricate interactions and ramifications of the global response to climate change. A particular emphasis is placed on examining climate and ocean-meteorological boundaries to gain insights into their impacts.

5. FungiDetect-Ensemble: A Novel Model for the Comprehensive Detection of Diseases in Tomato Leaves

   R. Usha, Radhika Baskar

   Abstract

   This study introduces a comprehensive methodology for detecting fungal infections on tomato leaves using advanced image processing and machine learning techniques. The procedure begins with the acquisition of high-resolution tomato leaf images, which are then resized to a standard dimension to ensure consistency and computational efficiency. The methodology includes several critical preprocessing steps, such as converting RGB images to grayscale to reduce computational complexity and emphasize intensity variations, as well as using the Canny edge detection algorithm to clearly distinguish leaf edges and features. To improve feature visibility and contrast, image enhancements such as adaptive thresholding and Contrast Limited Adaptive Histogram Equalization (CLAHE) are used. Images are segmented to isolate regions indicative of fungal infections using Gaussian Mixture Models, which can handle complex color and texture distributions effectively. Local Binary Patterns (LBP) are used for feature extraction because they are efficient at capturing textural patterns associated with leaf infections. To classify infection types and understand infection progression, a novel ensemble model, FungiDetect-Ensemble, is used, which combines the strengths of multiclass SVM models and Recurrent Neural Networks, specifically LSTM. This ensemble approach ensures a nuanced understanding of fungal infections by combining immediate classification and temporal analysis.

6. HOVG-Net: A Hybrid Model for the Early Detection of Lung Cancer in Computed Tomography

   Syed Zaheer Ahammed, Radhika Baskar, G. NalliniPriya

   Abstract

   This study presents a comprehensive method for detecting lung cancer in CT scans using the LIDC-IDRI dataset, advanced image processing, and deep learning. The procedure includes image resizing, noise reduction via median filtering, binary conversion via thresholding and refinement via morphological opening. The analysis begins with image segmentation using active contours, which is followed by feature detection using the innovative HOVG-Net model, which combines the Circular Hough Transform and VGG-19. The evaluation phase quantifies the presence of lung cancer by revealing 59 circular features covering a total area of 2916.3635. This methodology not only accurately identifies lung cancer indicators, but it also provides a clear visual representation of the findings, which contributes significantly to improved medical diagnosis and treatment planning.

7. Customer Satisfaction Prediction in Online Goods Delivery Through Interpretable Predictive Models and Sentiment Analysis

   Akula Venkata Satya Sai Gopinadh, S. V. S. N. Sarma, Gudipudi Radhesyam

   Abstract

   In this paper, we explore the effectiveness of machine learning (ML) in forecasting customer satisfaction scores based on the sales dataset curated from Olist, a prominent Brazilian e-commerce enterprise. Customer satisfaction score is classified into four distinct categories: Poor, Average, Good, and Excellent, with a prevalence of Excellent ratings among the majority of sales orders. Motivated by the recognition that delivery duration and product, seller rating scores, derived from previous customer transactions, play pivotal roles in shaping customer satisfaction, we embark on a comprehensive analysis. In our investigation, we leverage advanced machine learning techniques, specifically Random Forest (RF), XGBoost (XGB), and Decision Tree (DT) to forecast customer satisfaction scores. Additionally, we incorporated sentiment analysis of review comments into our research. Notably, the XGBoost (XGB) model emerged as the top performer, achieving an average precision, recall, and macro F1-score of 0.53,0.52 and 0.53 respectively. This underscores the effectiveness of incorporating sentiment analysis alongside traditional ML models in predicting customer satisfaction in e-commerce settings.

8. Comparative Analysis of Machine Learning Techniques for Cervical Cancer Prediction

   B. Sarada, A. Guru SSVS Murali Krishna, P. Sarayu Sree Yadav, K. Lakshmi Puspha, S. Revathi, Siva Sankar Namani

   Abstract

   Data mining, an interdisciplinary field bridging computer science as well as statistics, extracts insights using databases for the aid of decision-making. Within Data mining, classification involves learning from existing cases to predict the class of new cases. We investigate the performance of Decision Trees, ANN, KNN, Naive Bayes classifiers and SVM implemented in the Orange and WEKA platforms [1, 2]. Through comprehensive experimentation and evaluation, we assess the predictive capabilities and effectiveness of these ML techniques. The objective is to assist analysts in swiftly achieving effective results. This paper presents a thorough examination of accuracy calculation methodologies for assessing the performance of algorithms on data mining platforms.

9. Interpretable Machine Learning and Sentiment Analysis for Enhanced Predictive Accuracy in Financial Markets

   Saideva Sathvik Ravula, S. V. S. N. Sarma, Gudipudi Radhesyam

   Abstract

   In the realm of finance, where news sentiment can sway markets, our study explores the synergy between FinBERT, a specialized BERT variant trained for finance, and Ridge Regression, an interpretable machine learning model. As financial decisions hinge on both performance and transparency, we prioritize models that strike a balance, avoiding excessive complexity while delivering robust predictions. Leveraging Google News as our source, we harness its clustering capabilities to consolidate relevant articles. Through this fusion of FinBERT’s sentiment analysis and Ridge Regression, we offer actionable insights into stock market behavior while demystifying the decision-making process. Our research emphasizes the importance of maintaining transparency in financial predictions.

10. Enhancing Cyclone Intensity Prediction Using Deep Learning Models with INSAT- 3D IR Imagery

    Sireesha Vikkurty, Nagaratna P. Hegde, Sriperambuduri Vinay Kumar, Myada Tejaswini, Tanguturi Kranthi

    Abstract

    Tropical revolving storms also known as cyclones, hurricanes and or typhoons depending on the geographical location cyclones are severe natural calamities that pose a huge risk to coastal communities and marine shipping. Correct anticipation and measurement of cyclone intensity is important particularly in disaster management. Sometimes, the traditional method of calculating the intensity of cyclones depends more on the interpretation of satellite imagery which is time consuming and usually subjective and hence all the inaccuracies. However, in the recent past Deep learning based analysis for cyclones has proven itself to be very useful, making the process of automating and optimizing cyclone intensity prediction more effective. This review paper aims to review recently developed deep learning techniques in cyclone intensity prediction. The paper also details various analyses of the various deep learning architectures that have been used successfully in cyclone intensity prediction. Finally, the paper gives insight into the cyclone intensity prediction performance by various deep learning models, with comparison of their attributes, pros and cons.

11. Real Time Facial Emotion Detection Using Deep Learning

    R. Anusha, G. Roja, N. Divya, CH. Veena

    Abstract

    A real-time facial emotion detection system based on deep learning, specifically a CNN is presented in this study. The model is trained on a diverse dataset to recognize basic emotions, and it uses transfer learning to improve performance, especially with limited labeled data. To ensure usability, real-time processing optimization techniques are used, and the system is designed for deployment on standard hardware. Benchmark evaluations demonstrate the effectiveness of the proposed system, highlighting its potential applications in human-computer interaction, virtual reality, and emotion-aware systems. In addition to giving priority to accuracy, the system also emphasizes real-time efficiency, which is attained by applying optimization techniques. The suggested facial emotion detection system has the potential to be widely used because it is made to be installed on common hardware. The system’s contributions are expected to be crucial in forming emotionally intelligent technologies, revolutionizing human-computer interactions, and offering insightful information about user experiences in dynamic scenarios as the technological landscape develops.

12. Enhanced Vehicle Detection using Scalar Invariant Feature Transform

    Amara Rithik Raj, Ganesh B. Regulwar, Rangineni Anvitha, K. Venkatesh, Ashish Mahalle

    Abstract

    Unmanned aerial vehicles (UAVs) provide new possibilities for civilian remote sensing, including automated vehicle detection. This paper proposes a novel approach using Scalar Invariant Feature Transform (SIFT) for feature extraction. This process distinguishes vehicle-related key points from others through machine learning approaches, including SVM, CNN, YOLO, & SSD. Our real-world UAV experiments showcase effective vehicle detection. Vision-based vehicle identification faces challenges due to changing road conditions. In this paper present a cost-effective, real-time, accurate detection method. SVM classifiers with swift Haar-like features detect, while virtual detection lines mitigate false positives. SIFT-based classification improves accuracy and minimizes missed detections. Multi-class classification utilizes YOLO, SSD, CNN, and SVM. Our approach promises robust UAV-based vehicle detection and contributes to intelligent transportation systems.

13. Real Time Traffic Sign Detection Using Deep Neural Networks

    P. Indrani, Ganesh B. Regulwar, Md Sohel Ahmed, K. Sneha Reddy, Vaibhav, Mohd Salahuddin

    Abstract

    Street signs are the unsung heroes of our roadways, silently guiding and instructing drivers to ensure the safe and efficient movement of vehicles. However, despite their critical role, accidents frequently occur when drivers either fail to notice these signs or misinterpret their meaning. The proposed Traffic Sign Recognition and Alert System represents a significant leap forward in addressing these issues and elevating road safety to new heights. This innovative device leverages advanced technology to recognize traffic signs and promptly alert drivers through the vehicle's speakers, enabling them to make informed decisions and contribute to a safer road environment for all. The Convolutional Neural Network (CNN) is a deep learning technology built to excel at image identification tasks It is the brain of this ground-breaking system. The German Traffic Sign Benchmarks Dataset provided the dataset, which is considerable at 51,900 pictures of traffic signs.

14. Multimodal Demographic Prediction: A Transfer Learning Framework with EfficientNet Model

    Lalitha Gehlot, Arshanapally Pooja, E. Ravi Kumar, Manzoor Mohammad, Swathi Sambangi, Nikhila Kathirisetty

    Abstract

    The exploration of facial features has garnered substantial interest, especially with the evolution of deep learning methods. There are various uses for demographic characteristics like age, gender, and ethnicity, yet researchers haven't extensively investigated the prediction of ethnicity within this domain. This research delves into predicting multiple human attributes using the EfficientNet model in deep learning. Along with human race, the model also focuses on predicting age and gender. The well-regarded EfficientNet model, known for its adaptability and high performance, undergoes extensive training using a varied dataset inclusive of diverse ethnicities, age ranges, and gender. Specifically, we employ a variant of the EfficientNet architecture, B0, renowned for its adaptability and high performance. It covers the classification of human races, age estimation, and gender determination, utilizing the robust UTK Face Dataset sourced from Kaggle. Assessment metrics such as accuracy, Mean Absolute Error (MAE) are employed for this evaluation.

15. Genetic Prognosis: Harnessing DNA for Disease Prediction

    K. Sreeveda, Y. Rajyalaxmi, N. Divya, Karella Harshini, Kudurupaka Vaishnavi

    Abstract

    To understand a disease, the need to find out which genes are involved in causing it is crucial. Identifying and linking genes to diseases is difficult and costly to conduct experiments involving a huge pool of potential candidate genes. As a result, alternative computational methods that are both cost-effective and easy have been introduced for identifying candidate genes linked to specific diseases. Because genes causing the same or similar diseases have less variance in their sequence or network features of protein-protein interactions, the majority of these strategies rely on phenotypic similarities. This is based on the idea that genes are located closer together in the protein interaction network that causes the same or similar disease. Nevertheless, these techniques solely rely on fundamental network properties, topological features, gene sequences, or existing biological knowledge as a preliminary, limiting the identification process to individual gene-disease associations. The introduction of innovative computer-driven approaches to discover genes that play a role in various diseases is done. These approaches can be used as tools that use computational power to help understand which specific genes might be connected to different health conditions.

16. An Efficient Neovascularization Detection in Fundus Images Using Transfer Learning

    D. V. Lalita Parameswari, R. Pallavi Reddy, Aakifah Fatima

    Abstract

    Individuals with diabetes face the risk of developing Proliferative Diabetic Retinopathy (PDR) which is a retinal disorder that leads to neovascularization. Multiple studies propose image processing for neovascularization detection, but its unpredictable growth and small size still pose challenges. Deep learning techniques with automatic feature extraction are gaining prevalence in neovascularization detection. The proposed work introduces a method based on transfer learning. Two distinct models are constructed using transfer learning, one is utilizing MobileNet and the other is combining GoogLeNet and ResNet18. An alternative method involves pre-trained CNN for feature extraction followed by SVM classification. A comparison was conducted across all methods in terms of accuracy. MobileNet achieves the highest accuracy of 97.6%.

17. Analyzing Methodological Approaches for Mobile Sink Prediction and Trust-Aware Routing for IoT-Wireless Sensor Networks

    N. Reshma Chandrika, Leena Arya

    Abstract

    This paper elaborates on the Mobile Sink Prediction module for the Internet of Things-Wireless Sensor Networks (IoT-WSN). It proposes a learning mechanism, maybe involving Deep Long Short-Term Memory (LSTM), for forecasting mobile sink locations. Moreover, an optimization algorithm is used to tune system performance depending on the predicted sink locations, e.g., energy efficiency and network lifetime. The adaptive routing algorithms are developed with the help of evolutionary optimization and deep learning techniques to optimize data transmission paths based on the inferred sink locations and network conditions. The paper further examines the trust-aware model, underlining that trust management is key to network security and robust communication in wireless sensor networks. An evaluation of the performance of the trust-based routing algorithm is performed via this statistical demonstration, and throughput and residual energy are considered as this result is supplied for the information about the effectiveness of secure and energy-efficient data transmission in IoT-WSN systems.

18. Empowering Youth: The Role of AI Chatbots to Combat Depression, Anxiety and Address Mental Health Challenges

    Adithya Chintalapudi, V. S. S. S. Srujan Kuppambhotla, Satwik Mandalemula, Sruthi Kowdagani, Leena Arya

    Abstract

    Over the last decade, challenges relating to mental health, such as depression and anxiety among young people, have been frequent. However, access to traditional services for mental health often presents significant barriers for this demographic. This paper explores the emerging role of AI chatbots in providing accessible and personalized support to empower youth in managing their mental health challenges. By leveraging artificial intelligence, chatbots offer a unique opportunity to bridge Mental health disparities, care delivery, providing timely interventions, and reducing the stigma associated with seeking help. Through a comprehensive review of literature, case studies, and ethical considerations, this paper examines the potential benefits and challenges of integrating AI chatbots into youth services for mental health.