SOFTWARE EFFORT PREDICTION USING MACHINE LEARNING TECHNIQUES

Abstract

Precise estimation of software e ort is a crucial task in the software engineering domain. The e ort is the most important factor which a ects the budget of a project. Therefore, software e ort estimation is very crucial and there is continuously a necessity to improve its accuracy as much as possible. For a quality software e ort, both over-estimation as well as under-estimation may lead to very dangerous consequences. Therefore, it is very important to determine the best technique which can give exact results for software e ort estimation. In this study, we analyze several machine learning (ML) techniques like bagging, linear regression, KStar, M5Rules, RIP (Reduces Error Pruning) Tree and Multilayer Perceptron (MLP) in order to develop models to predict software e ort. Two di erent datasets i.e. China dataset and Albrecht dataset have been used in our research. Results of machine learning algorithms can be di erent from dataset to dataset. Multilayer perceptron has shown good performance for China dataset and REP Tree shown for Albrecht dataset.