Enhancing Software Bug Prediction with Extreme Learning Machine: An Empirical Analysis Using NASA Datasets

Predicting software bugs before development cycles allows developers to mitigate potential issues before they escalate into significant problems. This paper presents a novel hybrid approach for Software Bug Prediction (SBP), integrating Principal Component Analysis (PCA) and Linear Discriminant Analysis (LDA) for dimensionality reduction with an Extreme Learning Machine (ELM) classifier. The outlined scheme significantly enhances prediction accuracy by using PCA to minimize the feature space while maintaining 95% of the variance and LDA to optimize class separability. Tested on high-dimensional datasets, the model achieved a remarkable accuracy of 98.49% on the MC1 dataset, outperforming other leading models in the field. The results underscore the effectiveness of the PCA+LDA+ELM combination in capturing complex patterns in software bug data, leading to more accurate predictions. Furthermore, high scores in the recall, F1, and Matthews Correlation Coefficient (MCC) metrics demonstrate the model's robustness and applicability in real-world software engineering environments. This hybrid technique offers a reliable and efficient solution for improving software quality through early defect prediction.