The Qasir application is one of the Point of Sale (POS) platforms widely used by Micro, Small, and Medium Enterprises (MSMEs) in Indonesia to manage transactions and business operations. User reviews on the Google Play Store platform provide valuable input on system performance, feature reliability, and user satisfaction levels. This study aims to compare the performance of three machine learning classification algorithms, namely Naive Bayes, Support Vector Machine (SVM), and Random Forest, in classifying user review sentiment into positive and negative categories. The dataset consists of 2,500 reviews obtained through web scraping techniques from the Google Play Store. The pre-processing stage of data includes text cleaning, case folding, tokenization, stopword removal, stemming, and feature extraction using Term Frequency-Inverse Document Frequency (TF-IDF). The model evaluation was carried out using a 10-fold cross-validation technique based on the Accuracy, Precision, Recall, and F1-Score metrics. The results of the experiment showed that the Support Vector Machine (SVM) algorithm produced the highest performance with an accuracy of 85.20%, Precision of 84.80%, Recall of 85.10%, F1-Score of 84.95%, and a standard deviation of 1.25%, outperforming Random Forest (82.40% accuracy) and Naïve Bayes (76.60% accuracy). These results prove that TF-IDF-based SVM is very effective in the sentiment classification of Indonesian texts and can provide strategic recommendations for developers to improve the service quality of the Qasir application