Journal of Android and IOS Applications and Testing

Stacked Ensemble with Logistic Regression Meta-Learner: A Multi-Level Framework for Enhanced Medical Diagnosis

Satish Kumar Kalagotla — 2026-04-03

Background: Stacked generalization combines multiple base learners with a meta-learner to improve predictive performance.

Objective: This paper proposes a novel stacked ensemble that uniquely integrates five optimized SVM variants DT-SVM (missing value handling), Correlation-SVM (multicollinearity-aware), ABC-SVM (feature-optimized), GS-GA-SVM (parameter-optimized), and Standard SVM with logistic regression as meta-learner for breast cancer diagnosis. No prior study has combined this specific set of optimized variants in a single stacking framework.

Methods: The framework uses a two-level protocol: Level-1 trains base learners with 5-fold cross-validation to generate meta-features; Level-2 trains logistic regression on these features. Novel contributions include: (1) the first bias-variance decomposition analysis of stacking for medical diagnosis; (2) interpretable meta-learner coefficient analysis to rank base learners by clinical importance; and (3) rigorous cross-dataset validation across four medical benchmarks.

Results: The ensemble achieves 99.12% accuracy (AUC-ROC: 0.9982) on the Wisconsin dataset, outperforming individual base learners (avg. 95.8%) and bagging (98.76%). Novel bias-variance analysis reveals bias and variance reductions of 61.2% and 78.2% versus the standard SVM. Cross-dataset validation confirms generalizability: PIMA (89.23%), Hepatitis (90.12%), Mammographic (91.28%).

Conclusion: The proposed stacking framework achieves state-of-the-art performance with novel contributions in bias-variance decomposition, interpretable meta-learning, and cross-dataset validation, demonstrating significant clinical utility for diagnostic support systems.

Results are based on benchmark datasets; clinical validation is required for real-world deployment.

Skykash: Random Forest-Based Airfare Price Prediction System

Shashank Bisht — 2026-03-23

Airline ticket prices fluctuate dynamically due to multiple factors such as booking time, route demand, seasonal variations, and airline pricing strategies. This variability makes it challenging for travelers to determine the optimal time to purchase tickets. This study proposes Skykash, a web-based airfare price prediction system that leverages supervised machine learning techniques to forecast ticket prices accurately. The system evaluates three predictive models: Linear Regression, Random Forest, and Gradient Boosting using a dataset of 100,000 Indian flight booking records collected between 2023 and 2025. Model performance is assessed using Root Mean Square Error (RMSE) and Mean Absolute Error (MAE). Experimental results indicate that Gradient Boosting outperforms other models with the lowest RMSE (705) and MAE (510), demonstrating its effectiveness in capturing nonlinear pricing behavior. The predictive engine is integrated with a Flask backend and React-based user interface to provide interactive visualization and decision support. The proposed system assists travelers in identifying cost-effective booking periods and supports airline stakeholders in implementing data-driven pricing strategies.

Case Study on the Design and Object-Oriented Simulation of a Multi-Layer Intelligent Online Exam Proctoring Framework

Riya Nilesh Laddha — 2026-04-15

The widespread adoption of online examinations has raised critical concerns surrounding academic dishonesty, false accusations, and student privacy. This case study presents the design and object-oriented simulation of a multi-layer intelligent online exam proctoring framework developed for a mid-sized university facing increasing incidents of suspected misconduct and student dissatisfaction with automated monitoring systems. Rather than relying on machine learning algorithms or real-time biometric processing, the framework employs core object-oriented programming (OOP) constructs — including encapsulation, constructor overloading, recursion, static variables, and inner classes — to model intelligent decision-making behavior across five architectural layers: Candidate Management, Monitoring, Behavioral Analysis, Risk Assessment, and Administrative Decision. A cumulative, weighted risk-scoring approach replaces traditional binary event detection, thereby reducing false positives and enabling proportional enforcement. Three simulation scenarios validate the framework's logical consistency, scoring stability, and fairness across varying risk levels. Human-in-the-loop oversight is embedded as a governance mechanism to prevent full automation of disciplinary processes. The framework demonstrates that principled software architecture and OOP design can replicate intelligent monitoring behavior in a transparent, explainable, and ethically compliant manner, offering educational institutions a modular and extensible foundation for responsible digital assessment systems.

AI Assistant for Student Support (EduSupport AI)

Shivam Hingane — 2026-04-03

The AI-powered chatbot for student support is an intelligent virtual assistant designed to provide instant, accurate, and automated responses to student queries related to academic, administrative, and campus-related information. The system uses Artificial Intelligence (AI), Natural Language Processing (NLP), and machine learning techniques to understand user inputs, interpret intent, and generate relevant responses in real time. This chatbot enables students to access important information such as course details, exam schedules, assignment deadlines, admission procedures, and institutional policies without the need for direct human intervention. The chatbot is available 24/7, ensuring continuous support and reducing the dependency on faculty and administrative staff for routine queries. By automating repetitive tasks, the system helps reduce workload, improves operational efficiency, and enhances communication between students and educational institutions. Additionally, the chatbot learns from previous interactions and improves its accuracy and performance over time, providing more personalized and relevant responses. The implementation of an AI-powered chatbot improves student engagement, saves time, and ensures quick access to information, thereby enhancing the overall student experience. This solution supports digital transformation in education by integrating intelligent automation into student support systems. The proposed system is scalable, efficient, and capable of assisting educational institutions in providing reliable, accessible, and modern support services to students.

Sign Language Detection using Python

Naik D. C — 2026-01-22

Communication barriers faced by hearing-impaired and speech-impaired individuals remain a significant social challenge, particularly in daily interactions with non-sign-language users. Sign language serves as an effective medium for communication; however, the lack of widespread understanding limits its practical use. To address this issue, this project proposes a real-time sign language detection system using Python and deep learning techniques to automatically recognize hand gestures and convert them into readable text. The primary objective of the proposed system is to enable efficient and accurate interpretation of sign language gestures using commonly available hardware and open-source software tools. The system utilizes a webcam to capture real-time video streams, which are processed frame by frame using OpenCV. Image preprocessing techniques such as resizing, normalization, and noise reduction are applied to enhance image quality and improve detection performance under varying lighting and background conditions. A YOLO-based object detection model is employed to localize hand regions within each frame due to its high detection speed and suitability for real-time applications. The detected hand gestures are then classified into corresponding sign language symbols using a trained deep learning model. The recognized gestures are translated into meaningful textual output and displayed on the screen in real time, allowing seamless interaction between hearing-impaired users and normal users. The proposed approach demonstrates reliable performance with acceptable accuracy and low latency, making it suitable for real-world applications. Experimental observations indicate that the system can effectively recognize common hand gestures while maintaining smooth real-time processing on standard laptop configurations. Overall, the proposed sign language detection system offers a cost-effective, user-friendly, and socially impactful solution for improving accessibility and inclusivity. The system can be further extended by incorporating speech synthesis, supporting dynamic gestures, and expanding the dataset to include more sign language vocabularies for enhanced accuracy and scalability.