Particle Swarm Optimisation Based Feature Selection for Software Effort Prediction Using Supervised Machine Learning and Ensemble Methods: A Comparative Study

Abstract

An essential goal of software cost prediction is predicting accurate costs, which is a vital part of the software process. Individuals with various skills carry out most software development tasks, so efforts are measured by person-months and time. The overestimation and underestimation of developing costs is a future disappointment. Additionally, software venture managers will determine the cost of software development. The prevailing expense for any software is the cost of ascertaining efforts, making software cost estimation a crucial task. The past three decades have seen various approaches to software cost estimations based on software metrics. However, it is difficult to decide which model offers a more precise estimate on the data set. This article uses three different data sets: Albrecht, Desharnais and COCOMONASA. Classification often includes many features in a data set, but not all are useful for classification. Abusive and repetitive symptoms reduce performance, so this article uses particle swarm optimisation as a feature selection algorithm for different machine learning methods like Random Forest, REPTree, SMOReg, Linear Regression, M5Rule and MLP. At the same time, bagging is also used with said base learners. The bagging M5Rule gives the best performance with an MMRE value of 28.78%.