Comparative Analysis of Defect Prediction in Software through SqueezeNet and DenseNet Models

In software engineering, forecasting defects is a critical strategy for improving the strength and efficiency of software. This study undertakes a comparative analysis using datasets from NASA to assess the performance of two prominent convolutional neural network models, SqueezeNet and DenseNet, for predicting software Defect. SqueezeNet, known for its parameter efficiency, and DenseNet, celebrated for its feature propagation, were evaluated against various performance indicators such as accuracy, precision, recall, and the F1 score. The study presents a comprehensive examination of the models' performance on the task of binary classification of defective and non-defective software modules. Results indicate that SqueezeNet achieves remarkable precision and accuracy, making it particularly suitable for scenarios requiring the minimization of false positives. In contrast, DenseNet demonstrates robust recall abilities, favoring environments where the detection of all potential defects is critical. The insights gained from this analysis could aid in the strategic selection and implementation of neural network models for software defect prediction.