Machine learning based effort estimation using standardization

Abstract

Accurate estimations of software project effort is one of the most important tasks of software project development. The machine learning models have proven to provide high accuracy due to the learning natures of these techniques. Taking this into consideration, this paper aims at employing machine learning models of Random Forest (RF), Multilayer Perceptron (MLP) and Support Vector Machines (SVM) for purpose of predicting the effort. The Use Case Points (UCP) software size metric is used due to its advantage of providing estimation at initial stages of software development. The standardization preprocessing technique was applied on the dataset before training the models. The RF, MLP and SVM models were examined for the prediction accuracy. The experimental results obtained from RF model were better as compared to MLP and SVM.