<![CDATA[Software defects are one of the main causes of information technology waste, posing a major challenge in software development as they can degrade the quality of the software itself. To reduce costs and efforts in software development and maintenance, predicting software defects is the best approach. Multi-Layer Perceptron (MLP) is a type of artificial neural network that can be used to learn complex and non-linear patterns in input data. It excels in modeling complex and non-linear relationships in data, as well as automatically extracting features and handling problems that cannot be solved by linear models. One of the preprocessing steps to optimize MLP is data discretization, which involves dividing the range of attributes into intervals to reduce the number of numerical attributes to categorical data. Testing results with five types of data from NASA MDP (CM1, JM1, KC1, KC2, and PC1) showed significant accuracy improvements. In the CM1 dataset, accuracy increased to 96.1% compared to using MLP alone, which achieved 91.1%. In the JM1 dataset, accuracy increased to 79.1% compared to MLP alone, which achieved 78.3%. In the KC1 data, accuracy increased to 88.5% compared to MLP alone, which achieved 85.9%. In the KC2 dataset, MLP with discretization achieved an accuracy of 89.8%, better than MLP alone at 84.8%. In the PC1 data, the highest accuracy obtained was 95.5% compared to MLP alone, which achieved 94.3%.]]>
Copyrights © 2024
