Analysis of software fault prediction using machine learning algorithm

Abstract

Today software performs a requisite role in our daily lives. Software's complexity keeps growing. The increasing complexity of any software system making it very di cult to improve its quality. The performance of the software depends on its bugfree operation. The main goal of developing any software is to identify and resolve bugs that may be required in various situations before the schedule is established. Software fault prediction is a way that seeks to classify fault-prone software modules by using speci c underlying characteristics of software project before actual testing tends to start. Separate researchers have previously examined several classification ways for the prediction of software bugs. The output of various techniques varies from software to software, and no one technique is always successful throughout all elds. Nowadays, machine learning is widely using in software defect detection. We can save our valuable time and reduce costs by using machine learning algorithms in fault prediction. There are many machine learning algorithms used for the prediction of defects in software systems. Although most of the work is available for software systems classi cation, either fault-prone or non-fault prone, little attempt has been done to predict the fault ensemble techniques. We have set up a strategy in this paper to use some machine learning algorithms and Boosting Algorithms to analyze their performance on the promise dataset and uni ed Dataset. We have selected six machine learning algorithms, and they are KNN, Random Forest, Decision Tree, MLP, SVM, Nai ve Bayes, Logistic Regression and two Boosting Algorithms such as XGBoost and AdaBoost Algorithm. We applied those algorithms to our two types of datasets, such as the Uni ed Dataset and Promise Dataset (JM1, PC1, CM1). We have decided to analyze the best machine learning algorithm based on their maximum accuracy. We will ensure the best machine learning algorithm analysis for the uni ed and promise dataset.