A New Design for Software Defect Prediction based on Random Forest Classifier with Recurrent Neural Networks

Abstract

                Software systems have become increasingly complicated and adaptable in recent years. As a result, identifying and resolving software defects are critical. One of the most active study fields in software engineering is software fault prediction. As a result, software defect prediction is critical for improving software quality. It also aids in the reduction of software testing time and cost. As a result, many firms use it to predict software problems in order to save time, improve quality, and schedule resources to meet deadlines. The goal of software defect prediction is to forecast faults based on past data. As a result, it is difficult to anticipate in the real world because it necessitates a greater amount of data variables, metrics, and historical data. As software projects grow in scale, defect prediction techniques will play an increasingly essential role in assisting developers and reducing time to market with more dependable software products. The Random Forest with Recurrent Neural Networks (RF-RNN) techniques for forecasting software defects are introduced in this study. The performance of our suggested method for software defect prediction is measured using several characteristics such as True positive rate, false positive rate, precision, recall, accuracy, and execution time, which are all implemented on the MATLAB platform. Several existing strategies are compared to our proposed method. The graphical depiction of the comparative findings demonstrates that our proposed technique is a highly useful technique for forecasting software fault and provides improved prediction rates in a more efficient manner.