Kernel-based Class-specific Broad Learning System for software defect prediction

With the continuous expansion of the software industry, the problem of software defects is receiving more and more attention. There has been a series of machine learning methods applied to the field of software defect prediction (SDP) as a way to ensure the stability of software. However, SDP suffers from the imbalance problem. To solve this problem, we first propose a class-specific broad learning system (CSBLS), which assigns a specific penalty factor to each class in accordance with the data distribution. Then we design a class-specific kernel-based broad learning system (CSKBLS), which adopts kernel mapping instead of random projection. This additive kernel scheme takes into account both outliers and noise in the data set. Extensive experiments on the real-world NASA datasets show that CSKBLS outperforms the comparison methods on the tasks of software defect prediction.