Parameter Settings in Statistical Machine Learning and Evidential Neural Network Techniques for Software Defect Prediction: A Detailed Analysis

Abstract

This comprehensive study explores the significant influence of 
hyperparameter tuning on the effectiveness of statistical, machine 
learning (ML) and a newly evidential neural network models in Software 
Defect Prediction (SDP). The research investigates how optimizing key 
parameters such as learning rates, tree depths, and kernel functions 
enhances model robustness and predictive accuracy. By conducting a 
thorough analysis of parameter settings across various techniques, this 
study identifies the primary factors that contribute to improved model 
performance and reduced correlated metric issues. The findings 
underscore the critical need for systematic hyperparameter tuning to 
overcome model underperformance, enhance scalability, and ensure 
consistent predictive outcomes in diverse SDP environments. Emphasis 
is placed on the necessity for automated tuning mechanisms to sustain 
high levels of prediction accuracy and model efficiency. This study 
investigates the impact of hyperparameter tuning on statistical, 
machine learning (ML) and evidential neural network models for 
Software Defect Prediction (SDP). The research highlights how 
optimizing parameters such as learning rates, tree depths, and kernel 
functions improves model robustness. Key findings emphasize that 
systematic tuning reduces correlated metric issues and improves 
accuracy across SDP models.