Recommendation of Machine Learning Techniques for Software Effort Estimation using Multi-Criteria Decision Making

Abstract

 For the development of the software industry, Software Effort Estimation (SEE) is one of the essential tasks. Project managers can overcome budget and time overrun issues by accurately estimating a software project's development effort in the software life cycle. In prior studies, a variety of machine learning methods for SEE modeling were applied. The outcomes for various performance or accuracy measures are inconclusive. Therefore, a mechanism for assessing machine learning approaches for SEE modeling in the context of several contradictory accuracy measures is desperately needed. This study addresses selecting the most appropriate machine learning technique for SEE modeling as a Multi-Criteria Decision Making (MCDM) problem. The machine learning techniques are selected through a novel approach based on MCDM. In the proposed approach, three MCDM methods- Weighted Aggregated Sum Product Assessment (WASPAS), Technique for Order Preference by Similarity to Ideal Solution (TOPSIS), and VIseKriterijumska Optimizacija I Kompromisno Resenje (VIKOR) were applied to determine the ranking of machine learning techniques on SEE performance based on multiple conflicting accuracy measures. For validating the proposed method, an experimental study was conducted over three SEE datasets using ten machine-learning techniques and six performance measures. Based on MCDM rankings, Random Forest, Support Vector Regression, and Kstar are recommended as the most appropriate machine learning techniques for SEE modeling. The results show how effectively the suggested MCDM-based approach can be used to recommend the appropriate machine learning technique for SEE modeling while considering various competing accuracy or performance measures altogether.