Software Complexity Prediction Model: A Combined Machine Learning Approach

Abstract

The need for computers increased quickly. As a result, the program is utilized in a significant and intricate manner. More complex systems are being developed by software businesses. Additionally, customers expect great quality, but the market requires them to finish their assignment faster. Different measuring methods are employed by software firms. Some of these include customer feedback after it has been given to customers, software testing, and stakeholder input. The objective of this project is to use a combination of machine learning techniques to predict software bug states using the NASA MDP dataset. The research process considered data preprocessing methods and applied singular and combination machine learning algorithms. To create the model, the single classifiers were combined using the voting method. Accuracy, precision, and recall were used to evaluate the model's effectiveness, along with tenfold cross-validation. The promising result was recorded by a combination of J48 and SMO classifiers. Before attempting to test the software product, the researcher retrieved attribute data from the source code; the complexity of the software product will then be ascertained using the constructed model. The main contribution of this study is to improve software quality by incorporating a machine learning framework into the present software development life cycle between implementation and testing.