Software defect density prediction using grey system theory and fuzzy logic

Defect Density (DD) is a cornerstone metric in software quality assessment, influencing decisions across quality planning, testing strategies, and resource allocation. However, inherent uncertainties within software module data significantly impede accurate prediction of DD. This paper proposes a novel predictive framework that leverages the strengths of grey system theory and fuzzy logic to address data imprecision in software defect measurement. Our approach introduces a fuzzy grey relational similarity metric that capitalizes on the benefits of both fuzzy logic membership functions and grey relational analysis for similarity assessment. We perform a rigorous evaluation comparing our model with eight established machine learning algorithms across 28 publicly available software datasets. The proposed framework demonstrates superior performance, particularly in scenarios characterized by high sparsity levels in the DD variable. We further analyze the model's efficacy across varying data sparsity levels, categorizing datasets into four groups based on sparsity ratios. Our findings reveal the model's dominance over alternative prediction methods in datasets exhibiting high and very high sparsity. Notably, traditional techniques like multi-linear regression and multi-layer perceptron exhibit limitations in handling such challenges within this specific problem domain and data landscape. Conversely, ensemble learning techniques emerge as viable alternatives for datasets with lower sparsity levels.