Nowadays, public opinion toward the National Police's (POLRI) image is deteriorating. With the explosive growth of social media in Indonesia, opinions on POLRI-related present-day issues on Twitter easily go viral, influencing sentiments among individuals regarding Indonesian law enforcement. Negative sentiments, at some point, may lead to the undervaluation of law enforcement and the failure of the legal system. Therefore, sentiment analysis on Twitter is essential for gaining considerable insights into public views and attitudes on POLRI-related topics. This research is to determine the most effective approaches between Lexicon, a natural language processing method that relies on a corpus, and machine learning, which contains Naive-Bayes, Support Vector Machine (SVM), Random Forest, and Logistic Regression (LR). These approaches have differences in classification types: probability and linearity. To organize the research process, the Cross-Industry Standard Process for Data Mining (CRISP-DM) Framework, which comprises five data mining activities, was employed. The confusion matrix was used as the model performance measurement, with Naive-Bayes emerging as the best among all the tested models. Additionally, the subjects related to POLRI were developed using topic modeling, generating three topics: street police or police station, police acknowledgment in neighborhood activities, and the activity of contacting the police.