Wireless network, Signal processing, Communication
11
Scopus Publications
Scopus Publications
Region-Aware Genetic Feature Selection with Demographic Meta-Integration and Ensemble Learning for EEG epilepsy Seizure Detection B. Harish Goud, Madhu Patil, Miguel Villagómez-Galindo, Víctor Daniel Jiménez Macedo, Saiprasad Potharaju, Kiran Sree Pokkuluri, MVV Prasad Kantipudi Journal of Artificial Intelligence and Technology, 2026 Accurate electroencephalogram (EEG)-based seizure detection is important for early epilepsy diagnosis and timely intervention, yet existing methods often trade predictive performance for interpretability. Deep learning models can achieve high accuracy but function as black boxes, limiting clinical trust. Conventional machine learning models are more transparent, but they often ignore neurophysiological structure and patient-specific metadata, which can reduce performance. To address this gap, this study proposes an interpretable framework that combines region-aware genetic algorithm feature selection, demographic meta-integration, and ensemble learning. EEG channels are first grouped into neurophysiological regions, including frontal, central-parietal, temporal, and occipital areas. A genetic algorithm is then applied within each region to identify the most informative channels while preserving clinically meaningful brain topology. The selected EEG features are combined with patient demographic and clinical metadata, including age, gender, medication status, and seizure history, to create a compact feature vector. This feature set is used to train a soft-voting ensemble of RandomForest, ExtraTrees, and XGBoost classifiers. The framework was evaluated on a dataset of 50 drug-resistant epileptic patients and achieved 99.28% accuracy with a very low false-negative rate. In addition to strong predictive performance, the proposed method remains inherently interpretable by indicating which brain regions contribute to seizure prediction, making it suitable for practical clinical deployment.
A Hybrid Ensemble and Explainable AI Framework for Predictive Maintenance of Industrial Equipment Miguel Villagómez-Galindo, Manjula G, Jagadeesh B N, Madhu Patil, Saiprasad Potharaju, Achyutha Prasad N Journal of Artificial Intelligence and Technology, 2026 A modern industrial system with its critical machinery is very sensitive to unexpected equipment failure and may experience extensive operation interruption, danger to safety, and cost. The traditional maintenance approaches, reactive and preventative, lack intelligence and flexibility to make predictions of the failure based on real-time information, leading to failures that are expensive to fix or unnecessary maintenance. This paper proposes a hybrid ensemble predictive maintenance(PdM) framework to assist in overcoming these drawbacks by combining potent machine learning (ML) models as classifiers to PdM and SHapley Additive eXplanation (SHAP) framework to make decision-making PdM transparent and interpretable. The suggested approach is trained on actual industrial sensor data comprising multivariate time-series data such as temperature, vibration, voltage, and pressure measurements. Data are preprocessed in a powerful way with the removal of redundancy, label encoding, and scaling. The accuracy, precision, recall, F1-score, and analysis of the confusion matrix are used to evaluate each model. Strikingly, the three ensemble classifiers had 100 percent success in the detection of faults, with SHAP values having obvious key features dictating forecasts. The newness of this method is that it is potentially high-accuracy and interpretable at the same time, which is a respite from deeper or federated learning models, which are typically high-computational-load methods. The study adds a scalable, accurate, and explainable PdM framework that can be part of the new smart manufacturing.
Carelog: An AI-Powered Mobile Application for Skin Disease Detection and Diabetes Prediction Madhu Patil, Dhanyashree G, Jayashree H N, Roshini D, Yeshaswini R Proceedings of IEEE International Conference for Women in Innovation Technology and Entrepreneurship Icwite 2025, 2025 This study presents the development of a comprehensive mobile application built using Flutter, designed to address key health concerns through advanced features: skin disease detection and diabetes prediction. Leveraging state-of-the-art machine learning models integrated into the application, the skin disease detection module identifies potential dermatological conditions from user-uploaded images, ensuring quick and reliable assessments. The diabetes prediction feature uses user-provided health parameters, such as glucose levels, BMI, and age, to predict diabetes risk with high accuracy, aiding early detection and preventive care. This cross-platform solution combines accessibility, functionality, and user-centric design to offer a seamless and holistic health management experience. The application demonstrates the potential of combining AI and mobile technology to enhance healthcare accessibility and promote proactive well-being.
Securing the Digital Data using SFLA based Deep Convolutional Neural Network in Medical Network Environment G N Keshava Murthy, N Achyutha Prasad, Piyush Kumar Pareek, Pankaj Zanke, Madhu Patil 2nd IEEE International Conference on Data Science and Information System Icdsis 2024, 2024 As a result of the rapid development of multimedia in relation to network knowledge, accessing digital material has become relatively straightforward. Therefore, more emphasis on picture watermarking is required for IP protection. For this reason, many picture watermarking techniques were developed; nevertheless, they suffer from a lack of transparency and are not as sturdy. Multimedia copyright protection is very important in the field of digital watermarking. Digital picture watermarking is the process of secretly removing a watermark and inserting it into a carrier image. The success of digital watermarking in protecting digital data has attracted a lot of academic interest recently. There has been a lot of recent focus on using deep learning networks in tandem with wavelet-oriented strategies for picture watermarking. However, these benefits cannot be provided by traditional watermarking methods, which need manual methods for both extraction and embedding. In light of this problem, this research is driven to present a unique deep learning-based wavelet-based method for producing watermarked pictures with increased invisibility. A newly created Convolutional Neural Networks (CNN) model with medical pictures, and the deep features are retrieved utilising its fully connected layers. Next, the Shuffled frog-leaping algorithm (SFLA) is used to train the model, which ultimately results in higher quality classifications. The experimental results validate that the suggested model outperformed the baseline on a number of different measures, PSNR. In F $(228 \\times 344)$ dataset thevWatermark1 ratio the speckle noise reach as 0.9950 and then gaussian noise reach as 0.9924 and then salt and peper noise reaches as 0.9972 respectively. And then the Watermark2 ratio the speckle noise reach as 0.9950 and then gaussian noise reach as 0.9924 and then salt and peper noise reaches as 0.9972 respectively.
Analysis and Classification of Breast Cancer Disease Via Different Datasets and Classifier Models Ravi Kumar Barwal, Neeraj Raheja, B R Mohan, Yamuna U, Sai Sudha Gadde, Madhu Patil International Journal on Recent and Innovation Trends in Computing and Communication, 2023 Nowadays, Tumour is one of the important reasons of human death worldwide, producing about 9.6 million people in 2018. BC (breast cancer) is the common reason for cancer deaths in females. BC is a type of cancer that can be treated when detected early. The main motive of this analysis is to detect cancer early in life using ML (machine learning) techniques. The features of the people included in the WDBC (Wisconsin diagnostic breast cancer) and Coimbra BC datasets were classified by SVOF-KNN, KNN, and Naïve Bayes techniques. The pre-processing data phase was applied to the datasets before classification. After the data pre-processing steps, three classification methods were applied to the data. Specificity and Sensitivity rates were used to calculate the success of the techniques. As an outcome of the BC diagnosis classification, the SVOF-KNN technique was found with a 91 percent specificity rate and 90 percent sensitivity rate. When the outcomes attained from feature extraction and selection are calculated. It is seen that feature extraction, selection, and data pre-processing techniques improve the specificity and sensitivity rate of the detection system.
Design and development of frameworks for CPU verification efficiency improvement Sheetal Singrihalli Hemaraj, Shylashree Nagaraja, Sunitha Yariyur Narasimhaiah, Madhu Patil Indonesian Journal of Electrical Engineering and Computer Science, 2023 <p>Bug finding is a critical component of the verification flow and is resource intensive.In a typical week, a debug engineer writes triages, which take up significant amount of time that could be spent debugging another unique issue, and the lack of standardization in scripting causes maintainability issues in functional verification bug triage. A framework that allows customizable triage script generation is developed based on inputs from the engineer deploying YAML isn’t another markup language (YAML) files and practical extraction and report language (PERL) scripting, and this methodology is made automated and is standardized across projects to ensure maximum benefit going forward. The use of auto-triage in the project of functional verification bug triage has contributed to a 18% increase in triaged signatures on average, from 40% before its use to 58% after. A similar earlier project vs. current project comparison shows a 20% uplift. The triaged inputs that are parsed are currently being fed to a machine learning algorithm, which will help further improve the debug efficiency. As part of future work, the information from input YAML files can be used to analyze simulation failure attributes, hence improving the overall efficiency of debugging.</p>
Fetal Electrocardiogram Analysis Using Adaptive filters Manjula B M, Madhu Patil, Deekshith K S, Prateek P Deshapande Mysurucon 2022 2022 IEEE 2nd Mysore Sub Section International Conference, 2022 It is critical to obtain information about the fetus early in pregnancy to avert stillbirth. Medical personnel use Cardiotocography (CTG) to monitor the fetus's health in the hospital, however, it is not possible to record continuous long-duration signals using this method. As a result, constant and long-term monitoring of fetal electrocardiogram[1] signals is required to determine the health condition utilizing portable instruments. The invasive approach is superior to an invasive method for measuring ECG signals. To retrieve FECG encoded in the mother ECG, compact electronics and advanced signal processing techniques were required. Because the Fetal Heartbeat from the abdomen is frequently contaminated or interfered with by the Maternal Heartbeat, which is essentially noise. As a result, an attempt is made to separate the Fetal Heartbeat from the interfering Maternal Heartbeat in this case. The Adaptive [2] Noise Canceller (ANC) is used to remove the signal's noise content. Different adaptive filtering schemes, such as Single Input Single Output (SISO) on ANC, where adaptive algorithms such as least mean squares (LMS), Normalized least mean squares (NLMS), and leaky least mean [7] squares (L-LMS) are implemented in MATLAB and simulation results show the extracted FECG noise-free signal.
Energy efficient WSN by optimizing the packet failure in network Madhu Patil, Chirag Sharma Indonesian Journal of Electrical Engineering and Computer Science, 2017 <p>Wireless sensor network (WSN) has attained enormous growth in recent times due to availability of tiny and low cost sensor devices. The sensor network is been adopted by various organization for various application services such environment monitoring, surveillance etc.. The WSN are powered by batteries and are deployed in non-rechargeable remote location. Preserving batteries of these devices is most desired. Many methodologies have been proposed in recent time to improve the lifespan of sensor network among them clustering technique is the most sorted out technique. The drawback of existing technique the cluster head energy degrades very fast due to long transmission which requires amplification as a result energy is lost to the node that is surrounding the cluster head. They did not consider the packet failure likelihood among inter and intra as a results there exist scheduling bottleneck and degrades the energy of sensor devices. To overcome this work present a packet failure estimation model and hop selection optimization model for inter cluster transmission. Experiments are conducted for lifetime efficiency for varied sensor devices for proposed and existing. The result shows that the proposed model performs better than existing in term of network lifetime and energy efficiency.</p>
