Analysis of Customer Reviews of Fren.co Coffee & Eatery on Google Maps Using Logistic Regression and Random Forest Methods

Abstract

Online review platforms provide valuable data for evaluating customer perceptions and service quality in food and beverage businesses; however, such data are typically unstructured and frequently exhibit naturally imbalanced sentiment distributions that may influence classification outcomes. This study analyzes customer reviews of Fren.co Coffee & Eatery on Google Maps using Logistic Regression and Random Forest within a controlled comparative framework. A total of 225 valid textual reviews were collected and labeled into positive, neutral, and negative categories based on rating scores. The data were preprocessed through case normalization, cleansing, tokenization, stop word removal, and stemming, and subsequently transformed into numerical feature vectors using the Term Frequency–Inverse Document Frequency (TF-IDF) weighting scheme. To preserve the original sentiment distribution, an 80:20 stratified sampling strategy was implemented during model evaluation. Experimental results indicate that Logistic Regression achieved higher overall accuracy of 0.89 (89%) and demonstrated more balanced precision and recall across sentiment classes compared to Random Forest, which achieved an accuracy of 0.87 (87%) and showed stronger bias toward the majority class. These findings suggest that, in small-scale and naturally imbalanced Google Maps review datasets, linear classification models may provide more stable and consistent predictive performance than ensemble-based approaches. The study contributes empirical evidence on model behavior under realistic imbalance conditions and strengthens methodological understanding of classical machine learning applications for sentiment analysis in regional hospitality businesses.