Comparison of Hidden Markov Model with other Machine Learning Techniques in Software Defect Prediction

In the field of Software Engineering, defect prediction is the most comprehensive and active area of study. It determines which modules are prone to errors and requires rigorous testing. This allows the testing resources to be used effectively while still adhering to the limits. Our project aims to find the model which gives the best results in identifying the defects. We have used the most commonly used machine learning (ML) models K Nearest Neighbours (KNN), Support Vector Machine (SVM), Radial Bias Function (RBF), Decision Tree (DT), Naive Bayes (NB), and the least commonly used model Hidden Markov Model (HMM). The datasets are taken from the NASA and SOFTLAB repository. We have chosen six datasets from the NASA CM1, KC1, JM1, MC1, MW1, and PC1 and four from SOFTLAB AR1, AR3, AR4, and AR5. Principal Component Analysis (PCA) is used to reduce the dimensions of the dataset as there were upto 39 metrics. The best performing model is then determined based on various performance measures including accuracy, precision, F1 score, and recall. Six out of ten datasets showed that SVM performed better than other techniques. HMM performed the best on the AR4 dataset, which had the least data imbalance, and the worst on the AR5 dataset, which had the fewest instances.