Intelligent Software Fault Prediction Using Machine Learning and Neural Network Techniques

Pooja Singh, Saoud Sarwar, Kaveri Umesh Kadam

The investigative methodology includes a careful process that begins with data cleaning and feature selection. Uses 10-fold cross-validation, which means it basically takes 10ths of the data and tests it to see if it's consistent. A number of machine learning algorithms are evaluated, including Decision Tree, Naive Bayes, Support Vector Machine, Random Forest, and Multi-Layer Perceptron. The classic criteria accuracy, precision, recall and F1 score are used to evaluate all. When the models were completed, the results showed that models that use several techniques (e.g., Random Forest) and/or neural networks (e.g., Multi-Layer Perceptron) generally outperform simpler models. Among them, the Multi-Layer Perceptron is best suited for predictions, and the Random Forest is strong across all types of datasets. These discoveries indicate that machine learning can be used to estimate software failure time. Based on the research involved and the results it yielded, it is suggested that further investigation be undertaken into the possibility of combining the methods and emphasizing learning-based approaches. In this work, a methodology utilising data preprocessing, followed by feature selection and ten-fold cross-validation, is used to obtain a fair assessment. The performance of various classifiers, including Decision Tree, Naïve Bayes, Support Vector Machine (SVM), Random Forest, and Multi-Layer Perceptron (MLP), is assessed using four popular metrics: accuracy, precision, recall, and F1-score. The findings show that the ensemble and neural network outperform traditional classifiers for some data sets. The Multi-Layer Perceptron achieves the best predictive accuracy, while the Random Forest shows consistent and reliable performance. This suggests that machine learning has potential applications that go beyond simply predicting software failure in an early and efficient manner and explores further in the realm of hybrid and learning-focused approaches.