Classification of University IT Helpdesk Tickets Using Support Vector Machine with Hyperparameter Optimization

The classification of IT helpdesk tickets is crucial for improving response efficiency in service management systems, particularly within academic institutions. However, the process is still mostly manual, increasing the risk of misclassification. This study explores the use of the Support Vector Machine (SVM) algorithm with four kernel functions — RBF, Linear, Polynomial, and Sigmoid — to automate the classification of user-submitted service tickets. The dataset was sourced from the Telkom University service desk application database, covering 2023 and 2024, and comprises 13,508 records across nine service categories. Preprocessing steps such as stemming, stopword removal, and TF-IDF feature extraction were applied before model training and evaluation. The RBF kernel achieved the highest accuracy at 85.04%, followed by Linear at 80.64%, Sigmoid at 75.94%, and Polynomial at 63.69%. The internet access category had the best classification performance across all kernels, with RBF and Linear achieving F1-scores of 90% and 89%, respectively. The request data category also showed consistently strong results with F1-scores above 80%. Misclassifications were mainly due to overlapping vocabulary, data imbalance, and limited semantic variation in ticket descriptions. The results indicate that the RBF kernel is most suitable for this multi-class classification task. This study highlights the effectiveness of machine learning in improving helpdesk automation and provides a basis for future enhancements, such as incorporating semantic-rich features and addressing class imbalance. Notably, this research contributes a comparative analysis of different SVM kernel performances, which has not been extensively explored in previous research.