Prediction of Purchase Volume Coffee Shops in Surabaya Using Catboost with Leave-One-Out Cross Validation

Indonesia's coffee consumption grew from 265,000 tons in 2015 to 294,000 tons in 2020. Averaging 2% annual growth with a projected 368,000 tons by 2024. One of the coffee businesses is coffee shops, Coffee shop businesses often struggle to attract customers quickly, risking low purchase volume within their first five years. In their first year, challenges include management, company size, service quality, and customer preferences. This study adopts a quantitative approach and new solutions to develop a purchase prediction application based on machine learning and strategy to enhance purchase volumes for three coffee shops in Surabaya. It utilizes CatBoost, with LightGBM as a comparison, across multiple coffee shop locations. LOOCV (Leave-One-Out Cross-Validation) is used in this model to address research limitations, such as data overfitting and biases, while enhancing evaluation accuracy. As a result, the study established CatBoost as the superior model for purchase prediction, providing insights and practical applications in business forecasting. The Catboost model achieved an MAE of 0.91 and MAPE of 15%, outperforming LightGBM’s MAE of 1.13 and MAPE of 18%. These results confirmed CatBoost’s effectiveness for the coffee shop industry with good accuracy. This research also contributes to helping coffee shop owners in Surabaya understand market characteristics, such as the most profitable coffee types and high-customer-density locations. Additionally, it aids in optimizing purchase volume to leverage profit by developing new strategies based on prediction result. In conclusion, CatBoost accurately predicts purchase volume, helping coffee shops identify target markets and refine strategies based on customer preferences.