Enhancing Classification Algorithms with Metaheuristic Technique

Abstract

Classification is a process of grouping or placing data into appropriate categories or classes based on specific attributes or features to predict labels or classes of new data based on patterns observed from previously trained data. Implementing this process uses classification algorithms such as Naïve Bayes, Support Vector Machine, and Random Forest. However, the classification algorithm cannot classify data optimally due to the challenges in dealing with various data sets. Not all available features will make a solid contribution to the label of the data class, often in the form of noise or interference. For this reason, it is necessary to carry out a feature selection process. Currently, many feature selection processes have been carried out using correlation values from chi-square and gain-information, but the accuracy of the results is often still not good enough. This is because the chi-square and gain-information values are fixed. So, the selection of features is minimal and is not based on the previous learning process or what is known as heuristics. For this reason, in this research, several auxiliary algorithms are introduced to improve the performance of the classification algorithm, namely the meta-heuristic algorithm. Meta-heuristic algorithms are search techniques used to solve complex optimization problems, and these algorithms can help provide reasonable solutions in a shorter time than exact methods. In its operation, the metaheuristic algorithm optimizes the feature selection process, which will later be processed using the classification algorithm. Three (3) meta-heuristics were implemented, namely Genetic Algorithm, Particle Swarm Optimization, and Cuckoo Search Algorithm; the experiment was conducted, and the results were collected and analyzed. The result shows that combining Naive Bayes and Genetic Algorithm gives the best performance regarding higher accuracy improvement at +23.77%.