<![CDATA[This article discusses the implementation of the K-Nearest Neighbor (KNN) algorithm within a web-based recommendation system aimed at categorizing students from SMA Negeri 2 Tanjung Morawa into two distinct subject groups. The input data merges academic performance with insights regarding students’ interests and indicators of their abilities. Prior to classification, records of students and those of alumni with labels undergo Min-Max normalization, and the measurement of similarity is conducted through Euclidean Distance. Recommendations are generated with K set to 5, choosing the most commonly occurring label from the nearest neighbors. The system utilizes 128 labeled alumni records for training and analyzes 10 student records for testing. The results of the recommendation place 7 students in Subject Group 1 and 3 students in Subject Group 2. The effectiveness of the model is assessed by applying an 80:20 division to the alumni dataset, with the classification results being analyzed using a confusion matrix. The testing yields an accuracy of 80.00%, a precision of 67.86%, a recall of 65.00%, and an F1-score of 66.12%. Additionally, the evaluation section notes an average confidence of 98.08%, an average distance of 0.497, and confirms that all ten outputs fall within the high-confidence category. These results indicate that the KNN-based system is capable of offering data-driven assistance for recommendations on subject groups. However, enhancements are still necessary, especially regarding the enlargement and diversification of the training data as well as improving the attributes utilized within the recommendation model.]]>
