User reviews on e-commerce are one of the important elements in e-commerce. User reviews can help potential buyers make decisions based on the experiences and opinions of other people, for example women's e-commerce reviews. In providing positive, neutral or negative sentiment reviews, understanding customer perceptions is challenging. Classifying sentiment reviews will solve this problem, several classification techniques have been carried out, but there is still room for development in the use of simple machine learning techniques and sampling to overcome data class imbalance. Classification techniques used in this paper include Naive Bayes, SVM, and KNN. These algorithms will be compared to determine the most accurate model. Several preprocessing techniques are also carried out such to balance the dataset using ROS and SMOTE. It was obtained that the SVM method with ROS had the highest accuracy of around 0.94 for accuracy value, 0.93 for precision value, 0.94 for recall, and 0.92 for F1-score value. This research shows that the use of sampling techniques such as ROS and SMOTE can be effective in balancing imbalanced datasets, thereby improving model classification performance. These findings can be a reference for developing more efficient and accurate sentiment classification models, especially in the case of imbalanced data.