<![CDATA[The gender recognition system based on voice still faces challenges such as dependence on MFCC (Mel Frequency Cepstral Coefficients) features, which are not yet able to fully represent the complexity of human voice patterns. To overcome this, this study uses 20 voice characteristics and the K-Nearest Neighbor (KNN) algorithm because it is non-parametric, capable of handling non-linear relationships between features, and works intuitively by grouping data based on similarity of distance in the feature space, making it suitable for voice patterns that are not always linearly distributed. The purpose of this study is to analyze and develop a KNN model in classifying gender based on voice characteristics. Based on testing 50 variations of K values using K-Fold Cross Validation and Euclidean Distance, the evaluation results at K = 3, 5, and 7 showed average accuracies of 0.9740, 0.9700, and 0.9712. K = 3 was selected as the optimal parameter because it produced the highest accuracy. The results show that testing on 634 test data samples using K = 3 produced 619 correct predictions and 15 incorrect predictions, with an accuracy of 98% (0.9740), as well as precision, recall, and F1-score for the Female class of 0.98, 0.97, and 0.98, while for the Male class they were 0.97, 0.98, and 0.98.]]>
Copyrights © 2025
