Penerapan Naive Bayes untuk Klasifikasi E-Lapor Kabupaten Sumedang

Abstrak

Abstract: The E-Lapor (SP4N-LAPOR!) system in Sumedang Regency faces challenges in classifying the high volume of incoming public reports, where the manual disposition process to relevant agencies is slow and prone to error. This research aims to implement the Naive Bayes Classifier (NBC) algorithm with TF-IDF weighting to automate this classification. The methodology included text pre-processing (without stemming), an 80/20 data split, and evaluation using a confusion matrix. The results showed extremely low model performance, with a final accuracy of only 24.14%. The main finding of this study is that this poor performance was not caused by the NBC algorithm itself, but by fundamental issues within the dataset: an insufficient amount of data (data sparsity) with only ~145 clean samples, and a highly imbalanced class distribution. This finding concludes that the current E-Lapor Sumedang dataset is not feasible for automated model training without significant data augmentation and data balancing techniques. 

Keywords: E-Lapor, Text Classification, Naive Bayes, SP4N-LAPOR!, Imbalanced Data