MUMBAI, India, Aug. 12 -- Intellectual Property India has published a patent application (202641093563 A) filed by S R University on August 02, 2026, for An Ai Based Hybrid Sampling System And Method For Software Defect Prediction Using Imbalanced Datasets.
Inventors include Dr. Shaheen Layaq; and Dr. T. Sampath Kumar.
The application for the patent was published on August 07, 2026, under issue no. 32/2026.
Abstract: An AI Based Hybrid Sampling System and Method for Software Defect Prediction Using Imbalanced Datasets ABSTRACT The invention provides a hybrid sampling system and method that uses artificial intelligence to enhance software defect prediction in an imbalanced dataset. The framework consists of two parts: the first is the analysis of the characteristics of the dataset, such as class imbalance ratio, feature relevance, noise level, sample density, class overlap, and boundary distribution. Then an adaptive decision-making engine based on AI algorithms chooses appropriate oversampling and undersampling methods, schedules their order, and finally, optimizes the sampling ratios for each set of data. The region-aware oversampling method is used to create synthetic minority class samples only in safe and informative regions, and the confidence-based undersampling method is used to eliminate redundant majority class samples while retaining critical samples on the border regions. This balance data is then fed into machine learning classifiers like Random Forest, Support Vector Machine, XGBoost, LightGBM and neural networks. Several metrics are used to evaluate performance such as: precision, recall, F1-score, AUC, MCC, G-Mean, and balanced accuracy. The parameters of sampling are iteratively adjusted in a feedback optimization module to satisfy certain specified performance criteria. The system mitigates overfitting, loss of information, noise and bias in prediction, enhance robustness, scalability, generalization and early defect detection for various software repositories.
Disclaimer: Curated by HT Syndication.
