An Optimized Extreme Learning Machine Algorithm for Improving Software Maintainability Prediction

In recent years, software maintainability has become a critical attribute in software engineering to determine software quality. Hence, predicting this maintainability in an accurate and timely manner is a fundamental requirement for effective management during the software maintenance phase. This has led the software developers to pay more attention to those modules that need high maintenance. The current study proposes an Optimized Extreme Learning Machine (OELM) algorithm for Software Maintainability Prediction (SNIP) using three open-source datasets, viz.: Abdera, Ivy, & Rave. Since all these datasets are initially imbalanced, a Random Over Sampling technique is also used for re-sampling to avoid any problem encountered due to the imbalanced distribution of datasets. The predictive performance is analyzed based on the three performance evaluation measures, i.e., Accuracy, F1-Score, & Area wider the ROC Curve. The results support the effective utilization of the proposed OELM algorithm in SMP. The OELM algorithm's performance is also compared with four different Machine Learning (ML) algorithms, namely AdaBoost, Bagged CART, Flexible Discriminant Analysis, and Penalized Multinomial Regression. This comparison further supports the effectiveness of the OELM algorithm in predicting maintainability. The OELM algorithm performs 7.82%, 8.54%, and 22.98% better for Abdera, Ivy, and Rave datasets, respectively, than the other four ML algorithms taken together concerning Accuracy.