A study on gender detection using multiple classifiers on voice data

Researchers have frequently used metaheuristic algorithms for various problems due to their success. In data mining studies, feature selection (FS) is an essential preprocessing step for large-scale problems. Researchers have recently implemented FS using metaheuristic algorithms. In this study, the FS problem was solved using five different continuous metaheuristic algorithms (Osprey Optimization Algorithm, Spider Wasps Optimizer, Walrus Optimizer, Kepler Optimization Algorithm, and Crested Porcupine Optimizer) proposed in recent years. For the FS problem, the search spaces of continuous metaheuristic algorithms need to be converted to binary values. For this process, sixteen different types of transfer functions (S-shaped, V-shaped, Taper-shaped, and U-shaped) were analyzed. Comparison metrics such as fitness, accuracy, precision, recall, F1 score, number of selected features, and running time were used. The classification process was performed on the voice dataset consisting of 3168 samples and 22 features of male and female voices. K-Nearest Neighbor, Decision Tree, Random Forest, and Multi-Layer Perceptron were selected as classifiers. According to the mean fitness and accuracy results, the most successful classifier was determined to be K-Nearest Neighbor, and the most successful metaheuristic algorithm was determined to be the Kepler Optimization Algorithm.