Intelligent Systems

Inhaltsverzeichnis

1. Frontmatter

2. Effect of the Longitudinal Strain of PM Fiber on the Signal Group Velocity

   Karel Slavicek, David Grenar, Jiri Vavra, Martin Kyselak, Jan Radil, Jakub Frolka

   Abstract

   The polarization of the light can be used as the core principle of fiber optic sensors. One of the physical quantities which can be detected or measured this way is the longitudinal tension of the fiber. A set of measurements leading to approval of the suitability of polarization for this purpose was performed. This paper analyzes the dependency of differential group delay of the signal in slow and fast axes of the birefringent optical fiber on the longitudinal tension.

3. Machine Learning Algorithms Aided Disease Diagnosis and Prediction of Grape Leaf

   Priyanka Kaushik

   Abstract

   The range of diseases that can affect grape leaves has made it vital to analyze them. High-end data analytics and predictive analysis are required for a number of diseases, including black rot esca black measles, blight isariopsis, and others, in order to predict disease occurrence. For the prediction of leaf diseases, convolution neural networks combined with data augmentation have increased the degree of verification. For illness predictive analytics, a proper confusion matrix for support vector machines driven by CNN was created. Along with k-mean clustering, fuzzy logic with accurate feature extraction, and color moment definition, we also compared our results with these techniques. The findings indicate a higher effectiveness of up to 95% in correctly predicting grapes leaf disease.

4. Optimized Fuzzy PI Regulator for Frequency Regulation of Distributed Power System

   Smrutiranjan Nayak, Subhransu Sekhar Dash, Sanjeeb Kumar Kar, Ananta Kumar Sahoo, Ashwin Kumar Sahoo

   Abstract

   In this article improved fuzzy PI regulator is stated for frequency regulation of Automatic Control of distributed power systems. Originally, a two-region nonwarm framework is utilized. The advantage of the stated fuzzy PI regulator is shown with the help of contrasting the outputs. All real structure shows non-straight nature, subsequently, traditional regulators are not generally ready to give great and precise outcomes. So fuzzy-logic controller can be utilized to get more exact outcomes. The primacy of the stated hybrid particle swarm optimization & pattern search (hPSO-PS) approach adjusted fuzzy-PI selector over PS changed fuzzy PI selector, PSO changed fuzzy PI selector, hBFOA-PS changed PI selector, Differential Evolution (DE) changed PI selector and Bacteria Foraging optimization algorithm (BFOA) adjusted PI selector is demonstrated. It is seen that the Fuzzy PI regulator is more effective for controlling frequency relative to the PI regulator.

5. Detecting Depression Using Quality-of-Life Attributes with Machine Learning Techniques

   J. Premalatha, S. Aswin, D. JaiHari, K. Karamchand Subash

   Abstract

   Worldwide, depression affects millions of individuals even without their knowledge and is a crippling affliction. Primary care physicians frequently discover that they must treat mental health problems like depression despite having little or no formal training in how to do so. There is proof that an integrated strategy, where doctors regularly screen patients for mental health issues and collaborate with psychologists and other mental health specialists to treat patients, results in lower costs and improved patient outcomes. In order to handle and study the heterogeneous data and understand the correlation between aspects of quality of life and depression, this paper uses machine learning techniques. Machine learning is used to predict people who might have depression based on data that is found in CDC National Health and Examination Survey (NHAES) website. These forecasts could be used to more quickly and easily connect patients with qualified mental health specialists.

6. Patient Satisfaction Through Interpretable Machine Learning Approach

   S. Anandamurugan, P. Jayaprakash, S. Mounika, R. Narendranath

   Abstract

   In Patient satisfaction, the most important factor in assessing the quality is patient happiness. The happiness key factor impacts the health policy decisions. An individual’s specific health requirements, individualised treatment, and desired health results are of the utmost importance in the period of patient-centered care. Across the past decade, treatment delivery, management, and reimbursement practices have all been impacted by patient satisfaction as a clear insight and quality management of patient experiences. Using machine learning algorithms, the most relevant factors for patient satisfaction are founds.

7. Predicting the Thyroid Disease Using Machine Learning Techniques

   Lalitha Krishnasamy, M. Aparnaa, G. Deepa Prabha, T. Kavya

   Abstract

   An endocrine gland that is allocated in the front of the neck is called the thyroid, which produces thyroid hormones as its main job. Thyroid hormone may be produced insufficiently or excessively as a result of its potential malfunction. There are various thyroid types including Hyperthyroidism, Hypothyroidism, Thyroid Cancer Thyroiditis, swelling of the thyroid. A goiter is an enlarged thyroid gland. When your thyroid gland produces more thyroid hormones than your body requires, you have hyperthyroidism. When the thyroid gland in our body doesn’t provide enough thyroid hormones, then our body has hypothyroidism; when you have euthyroid sick, your thyroid function tests during critical illness taken in an inpatient or intensive care setting show alterations. Hypothyroid, hyperthyroid, and euthyroid conditions are expected from these thyroid conditions. The Three similarly used machine learning algorithms are: Support Vector Machine (SVM), Logistic Regression, and Random Forest methods, were evaluated from among the various machine learning techniques to forecast and evaluate their performance in terms of accuracy. Random forest can perform both regression and classification tasks. Logistic Regression is used to calculate or predict the probability of a binary (yes/no) event occurring. SVM classifiers offers great accuracy and work well with high dimensional space. A thyroid data set from Kaggle is used for this. This study has demonstrated the use of SVM, logistic regression, and random forest as classification tools, as well as the understanding of how to forecast thyroid disease.

8. An Automatic Traffic Sign Recognition and Classification Model Using Neural Networks

   Rajalaxmi Padhy, Alisha Samal, Sanjit Kumar Dash, Jibitesh Mishra

   Abstract

   The significance of traffic symbol recognition technologies, which have played a key role in street security, has been the subject of much interest to researchers. To accomplish their assessment, specialists employed Artificial Intelligence, deep learning, and image processing tools. Convolutional Neural Networks (CNN) are deep learning-based designs that have sparked a new and ongoing research into traffic symbol classifications and recognition frameworks. The objective of this paper is to establish a CNN model that is suitable for insertion purposes and has a high level of order exactness. For the series of street symbols, we used an upgraded LeNet-5 model. The German Traffic Sign Recognition Benchmark (GTSRB) information base will function as the framework for our model architecture, which outperformed existing models. GTSRB will have 99.84 percent accuracy. We decided to use a camera to verify the proposed model for an implanted application because of its softness and reduced number of boundaries (0.38 million) based on the improved LeNet-5 structure. The outcomes are advantageous, demonstrating the effectiveness of the discussed strategy.

9. An Artificial Intelligence Enabled Model to Minimize Corona Virus Variant Infection Spreading

   Dipti Dash, Isham Panigrahi, Prasant Kumar Pattnaik

   Abstract

   Many nations including India are being very badly affected by the second wave of the COVID-19 infections. The critical situation prevails in some states and cities of India. The mortality rate varies state to state depending on the health care facilities, immunological response of the individuals & comorbidities and vaccination status of that particular state. The multiclass prediction model is developed based on the status of data available from the different states of India considering their level of population density, intensity economic activities, education level, vaccination status and timing of lockdown or shut down. Based on this prediction model we can develop an application to motivate the internet of health things (IoHT), which can monitor the state and help in governing. This paper uses a multi class prediction model using Deep Neural Network (DNN) and validates the data set up to the year 2022, with accuracy level 98%. In this architecture, we have used 4 hidden layers between input and output layer. We have collected data from JHU CSSE Covid-19 and also follow our own algorithm to create our own dataset. We have taken 80% of data for training purposes and 20% of the dataset as validation purposes.

10. SoundMind: A Machine Learning and Web-Based Application for Depression Detection and Cure

    Madhusha Shete, Chaitaya Sardey, Siddharth Bhorge

    Abstract

    This paper presents a machine learning and web-based application for the detection of depression. The system mainly serves two components: two machine-learning-based models to detect depression and a web-based application. The first machine learning model is implemented to classify the positive and negative text entered by the user/patient. The negative text states the use of words indicating depression, which can be termed as one factor in deciding a patient's mental health. The model is built using libraries such as Natural Language Toolkit (NLTK), and WordCloud. The second model predicts the presence of depression based on multiple health-related features such as the patient’s data related to various other disorders he/she is having, age, weight, BMI, blood-related features such as levels of calcium, CO2, phosphorus, iron, etc., and work-life related parameters. The prediction is carried out based on the classification result implemented using Logistic Regression. The model predicts the results with 91.85% test accuracy, 93% precision, 99% recall, and 96% f1 score. The above-mentioned models are deployed on the web application. The web application not only helps in predicting mental health but also suggests the proper treatment to cure the condition.

11. Japanese Encephalitis Symptom Prediction Using Machine Learning Algorithm

    Piyush Ranjan, Sushruta Mishra, Tridiv Swain, Kshira Sagar Sahoo

    Abstract

    In India Japanese Encephalitis (JEV) has been a major public health problem. In endemic districts of country each year there is a large-scale outbreak occurring of JEV. Research says that Japanese Encephalitis is a flavivirus related to West Nile Virus, Yellow Fever and Dengue and it is escalated by mosquitoes. Japanese Encephalitis is although rare, but the fatality rate is around 30%. Till now there is no cure for JEV, the entire treatment is focused for supporting the patient to overcome disease and relieving severe clinical sign. Maximum number of JEV cases in India are of infants and the fatality rate is around 30% which is a great matter of concern. Here Force of Infection denotes the rate at which sensitive individuals acquire an infectious disease. In India, states which report major outbreak of Japanese Encephalitis are Uttar pradesh, Andhra Pradesh, West Bengal, Karnataka, Assam, Tamil Nadu, Bihar, Goa and Manipur. The impacting factors include Climate, Rice Distribution, Livestock Distribution, Population Density, Specific Age Group Density, Urban/Rural Category and Elevation. Impacting Factors may change with the location. Here we have used Machine learning algorithms like Ridge Regression, Lasso Regression, ElasticNet Regression and Multi-layer Perceptron for the prediction of Force of Infection of Japanese Encephalitis Virus. ElasticNet Regression Algorithm is also used for extracting the significant attribute from the JEV Dataset. The proposed model generated an optimum performance in context to the error rate and accuracy of prediction.

12. Smart Skin-Proto: A Mobile Skin Disorders Recognizer Model

    Sushruta Mishra, Shubham Suman, Aritra Nandi, Smaraki Bhaktisudha, Kshira Sagar Sahoo

    Abstract

    With the advancement and rapid development of the internet, the most convenient strategies for patients are mainly provided with digital healthcare systems that mainly includes the use of mobile health technology which is quite efficient. Moreover, this field is slightly shifting and also indicating interest towards the smart and intelligent models as there are quite a lot of benefits associated with it like cost decrement, easy to understand and also including the personal satisfaction of patients. The latest application of m-health medical treatment is now still on the process of the investigation because still users are facing challenges in the clinical environment. This m-health approach can be applied to accurately determine skin cancer symptoms in patients. In this paper, an impact of m-healthcare on disease diagnosis is demonstrated. A new m-health module for skin cancer diagnosis called ‘Smart Skin-Proto’ is developed. Then its usage in skin cancer assessment is also highlighted and upon implementation, the model records optimal performance which records an accuracy of 96.2% with 15 decision trees count. Also the overall latency of this application is less than other existing mobile apps.

13. Machine Learning Approach Using Artificial Neural Networks to Detect Malicious Nodes in IoT Networks

    Kazi Kutubuddin Sayyad Liyakat

    Abstract

    Devices can now effortlessly and wirelessly share data with one another over the internet or other networked systems thanks to a relatively new technology called Internet of Things (IoT). Despite these advantages, IoT systems are now more vulnerable to hacker attacks, which could lead to unfavourable outcomes. This is because of the IoT ecosystem’s continual expansion. These incursions may cause potential financial and physical harm. The Internet of Things is the automatically configuring network. This network is susceptible to a variety of attacks, all of which can be started by rogue nodes. For instance, during a denial of service attack, a malicious node bombards a targeted node with a large number of packets. For the purpose of locating these malicious nodes in a network, a threshold-based procedure utilising cutting-edge machine learning techniques is launched. By checking the path latency and alerting on it if it exceeds a set threshold value, the suggested method can help identify an attacker node. The NS2 programme will be used to mimic the suggested method. We evaluate the suggested methodology and demonstrate that our system performs well in terms of a number of measures, such as throughput, latency, and packet loss.

14. Real Time Air-Writing and Recognition of Tamil Alphabets Using Deep Learning

    S. Preethi, T. Meeradevi, K. Mohammed Kaif, S. Hema, M. Monikraj

    Abstract

    Writing has always been a prominent way of communication. The way in which the letters are written has been varying with time. From the conventional pen and paper to touch pad and stylus, the way of writing has evolved. Air- Writing is another development in which the characters are written in free space without being limited to a specific tool. This method of writing makes the hand movement easier compared to the conventional methods. Therefore, the air writing and recognition model will be of great help for children who start learning a language. The trajectory of the air written characters is obtained by mapping the focal point using Optical flow in OpenCV. The obtained trajectory is then preprocessed and given to Dense Net 121 which is a type of CNN model widely used for pattern matching along with the dataset from HP labs which contains 3000 images for 11 Tamil vowels. The model which is trained obtained a maximum training and validation accuracy of 98.2% and 91.83% respectively with minimum training and validation loss of 6.35% and 21.04% respectively.

15. A Fuzzy Logic Based Trust Evaluation Model for IoT

    Rabindra Patel, Sasmita Acharya

    Abstract

    Internet of Things (IoT) is a way of connecting the physical world to the internet where various devices are capable of communicating with each other. They all need a secure environment but the problem is implementation of any security approach for an IoT node is very difficult as resources of IoT nodes are very limited. So, trust evaluation and trust assessment are very important for IoT nodes. The paper proposes a fuzzy logic based trust evaluation model for IoT that employs different trust factors like End-to-end Packet Forwarding Ratio (EPFR), Amount of Energy Conversion (AEC), Packet Delivery Ratio (PDR) and Security Grade (SG) in order to construct a Fuzzy Inference System (FIS) that can calculate the trust value of each IoT node. Based on the resultant trust value, the IoT nodes are classified into three categories: Not Trustworthy, Not Sure and Trustworthy. An IoT node which belongs to Trustworthy only gets the access to forward the data packets or communicate with other IoT nodes. The Not Trustworthy and Not Sure IoT nodes are set to sleep mode for power conservation.

16. Supervised Learning Approaches on the Prediction of Diabetic Disease in Healthcare

    Riyam Patel, Borra Sivaiah, Punyaban Patel, Bibhudatta Sahoo

    Abstract

    There are many chronic diseases out of which Diabetes is one; that increases sugar level in the blood and is one of the most fatal that effect different organs in the human body. Diabetes can cause a variety of slow bad consequences if not detected and left without given medical care. The emergence of machine learning approaches, on the other hand, solves this crucial issue. The purpose and objectives of this work is to build a prototypical model that can properly forecast diabetes whether or not a person will suffer from it. To detect diabetes at an early stage, our work employs three classification algorithms based on supervised learning: Random Forest, Naïve Bayes Classifier and Multilayer Perceptron Network. The PIDD Database has been used in the experiments. The Precision, Accuracy, Recall, F-Measure, and ROC Area are all used to calculate the efficiency of the above three algorithms. The correctness and accuracy of a classification system is measured by the number of occurrences that are correctly classified and those that are mistakenly classified.

17. Solar Powered Smart Home Automation and Smart Health Monitoring with IoT

    Atif Afroz, Sephali Shradha Khamari, Ranjan Kumar Behera

    Abstract

    In this paper we present the prototype of smart home which is powered by solar. It has a smart MPPT, smart health care tracking system and a smart home automation system. The sensors are spread all across the entrance Gate, corridor, room and kitchen. This (IOT) design prototype has LCD transistor which keep on provides the information. We have also use Wi-Fi technology for online control and monitoring. we also have an LCD which keeps us providing the information regarding the data. We have also used Wi-Fi technology for the purpose of real time controlling and monitoring. The designed smart home utilizes the power from the solar panels through a maximum power point tracker and it has an LCD display which continuously gives us information regarding the solar input, charging efficiency and discharging rate etc. The internal infrastructure is so designed that it can work against an unexpected condition which may occur when the owner is not present in home and it also notify the owner about the problem that has occurred. All the power requirements of smart homes are met by a self-generated solar power.

18. Seasonal-Wise Occupational Accident Analysis Using Deep Learning Paradigms

    N. Nandhini, A. Anitha

    Abstract

    In recent years, occupational accidents causes a huge loss of human life and the development of the economy of the country. Many techniques are evolved for automating the safety precautions for employees in the industrial sectors such as mining, metals, construction, chemical, and electrical sections. However, the automation cannot be accurate as the data analysis is based on real-life data. Since the real-life data are imbalanced and uncertain, it is necessary to identify better tools to overcome these issues. Thus the proposed model utilizes SMOTE (Synthetic Minority Over-sampling Technique) for data balancing, whereas a rough set is used for identifying the significant features that help to maintain data consistency. The consistent data is then applied to the Deep Neural Network (DNN) for the classification process. The performance of the proposed model is checked against the evaluation metrics and compared with the existing deep learning models to exhibit the efficiency of the proposed model. Thus the findings of the proposed model may improve the abilities of safety professionals in the industrial sector to develop safety intervention activities.