Software Defect Prediction in Large Space Systems through Hybrid Feature Selection and Classification

Data mining and machine learning techniques have been used in several scientific applications including software fault predictions in large space systems. State-of the-art research revealed that existing space systems succumb to enigmatic software faults leading to critical loss of life and capital. This article presents a novel approach to solve this issue of overlooking software faults by utilizing both features selection and classification techniques to accurately predict software defects in aerospace systems. The main objective was to identify the preeminent feature selection and prediction technique that enhanced the software fault prediction accuracy with the optimal set of features. The investigations affirmed that a novel hybrid feature selection method revealed the most optimal set of predictive features although no particular predictive technique was suitable to predict faults in all space system datasets. Besides, the exploration of data mining techniques in fault prediction on the NASA Lunar space system software data clearly portrayed the improved fault prediction accuracy (~82% to ~98%) with the feature set selected by the proposed Hybrid Feature Selection method. Also, the random sub sampling method revealed an improved mean Matthew's Correlation Coefficient (MCC) and accuracy ranging from ~0.7 to ~0.9 and ~86% to ~98% respectively. This we believe generates further scope for future investigations on the most contributing space system features for fault prediction thus enabling design of aerospace systems with minimal faults and enhanced performance.