Improving Software Defect Prediction Accuracy through Modified Entropy Calculation in the Random Forest Algorithm

Abstract

Assume the scenario in which you are attempting to categorize software defects for a broad dataset. Which algorithm do you think will be the most effective for accomplishing that? Random Forest, Support Vector Machine, Neural Networks, Naive Bayes, K-Nearest Neighbours, Decision Tree, Logistic Regression, and other techniques are among those that can be utilized to solve the problem described above. The Random Forest technique, which allows for the generation of predictions through the utilization of many Decision Trees, is one of the most often utilized methods. Entropy, a complicated computation that examines the degree of uncertainty in the data, is the foundation upon which this algorithm is built. It is possible that the calculation of entropy, which is a function that uses natural logarithm, will take a significant amount of time. Does one know of a more accurate method for calculating entropy? The Taylor series expression was utilized in this investigation to investigate a different approach to calculating the natural logarithm. Any function may be approximated by utilizing its derivatives, and this series, which is made up of the sum of infinite terms, is what it is. Also, we updated the Random Forest algorithm by substituting the natural logarithm with the Taylor series equation in the Entropy calculation. This was done in order to achieve these modifications. Following the implementation of our improved algorithm on the dataset, we examined its performance in comparison to the Entropy formula that was first developed. An improvement in the algorithm's accuracy in predicting software defects was discovered by us as a result of our update to the technique