Performance Comparison of SVM Kernel Functions for Date Fruit Classification

In this study, the Support Vector Machines (SVM) algorithm was employed to classify different types of date fruits. The performances of various SVM kernel functions - namely Linear, Quadratic, Cubic, Medium Gaussian, and Coarse Gaussian- were compared during the classification process. The analyses were conducted using the Date Fruit Dataset, which was published on the Kaggle platform and comprises 34 numerical features. The Minimum Redundancy Maximum Relevance (MRMR) feature selection algorithm was utilized to identify the 13 most effective features for classification. Subsequently, classification was performed using both the complete feature set (34 features) and the selected subset (13 features). The findings revealed that the highest classification accuracy was achieved with the Linear kernel SVM model in both cases. When all features were used, the Linear SVM model reached an accuracy of 91.79%, whereas the accuracy increased to 92.07% when the 13 features selected by MRMR were employed. These results indicate that feature selection plays a significant role in improving classification performance.