Software Defect Severity Level Prediction Using Machine Learning Techniques

Abstract

Software Defects Are Errors That Prevent A Software Product To Function Properly; They Have A Huge Impact On Software Quality And Reliability. Identifying And Fixing Defects Early Will Result In Better Quality And Is Cost-Effective. Thus, Software Defect Managing Approaches Have Gained Much Attention In The Arena Of Software Engineering. After Defects Are Detected Knowing Their Magnitude Is Very Important, The Severity Level Of Defects Shows The Effect The Software Defect Can Cause To The End-User, Knowing The Stage Of Defect Is Now Primary Interest. Defects With The Highest Severity Level Should Have To Be Maintained Before Defects With Lower Severity, It Helps To Assign More Resources To The Most Defective Part Of The Software. Various Research Works Have Been Done In Defect Prediction, But It?�?S Also Important To Know The Stage Of Defects. Some Works Suggested Severity Level Prediction For Defect Reports From Bug Tracking Software. But None Of The Studies Addressed The Severity Level For The Software Module.This Study Proposed Software Defect Severity Level Prediction For Software Modules Using NASA?�?S MDP Repository Dataset That Contains Software Defect Metrics Such As Mccabe Metrics, Halstead Metrics, And LOC Metrics Which Determines The Defectiveness Of Module. The Study Used A Clustering Machine Learning Algorithm To Label The Datasets To Appropriate Classes Of Severity Level. Then After It Used RF, DT, KNN, And SVM Machine Learning Classification Algorithm To Perform The Classification Task, The Algorithms Are Applied To Five Datasets From The Repository Named CM1, KC1, KC3, PC, And PC3 To Develop A Model, Theperformance Of Each Algorithm Is Measured And Compared To Select A Better Model. Based On The Accuracy And F Measure, The Result Shows That The RF Model Performed Well On Each Dataset Compared To The DT, KNN, And SVM Models Which Resulted In Maximum Testing Accuracy Of 97.4% On PC1 And A Minimum Testing Accuracy Of 90.1% On KC1 .Datasets Also The F Measure Is 97% On PC1 And 90% On KC1. Generally, The Result Of The Study Shows Software Defect Metrics Can Determine The Severity Level Of A Defective Module And RF Model Can Be Recommended For Software Defect Predictions.