SOFTWARE DEFECT PREDICTION USING DYNAMIC FILTERING WITH ADAPTIVE NOISE-ENTROPY ALGORITHM

In software design, software defect prediction is a very important activity because this assists us in identifying at early stages, those modules that contain errors and thus improve the quality of software and lower maintenance costs. Combining Noise-Enhanced Dynamic Filtering (NEDF) as preprocessing and Adaptive Noise-Entropy Feature Selection (ANEFS) as feature selection, generates new approach delivered in this work. In order to enhance the data quality, NEDF technique removes data noise, outliers and software measure distortions using dynamic filtering, smoothing and transformation techniques. Through this approach, ANEFS facilitates effective reduction of features by adaptive balancing of relevance and redundancy of features and entropy-based feature evaluation and consequently removing the least informative features. These are objectively chosen traits directly responsible for improving the efficiency of software defect prediction by basing the models on the most discriminative trait, thus leading to correct identification of faulty modules not causing the confusion caused by irrelevant or redundant measures. By combining the two, the issue of the existing preprocessing and feature selection plans can be circumvented because it provides improved input data and more indicative features. This built-in structure maximizes the potential of software defect prediction consistency and predictability to a large degree, allowing for more realistic decisions about which modules are erroneous. Overall, this study shows that noise-resistant preprocessing and entropy-based adaptive feature selection is a promising avenue of research trying to improve the software defect prediction model.