Implementasi Metode Random Forest Untuk Prediksi Omzet Di Apotek Mitra Medical

Apotek Mitra Medical faces challenges in planning business strategies and operational management due to daily revenue fluctuations that are difficult to estimate manually. This uncertainty is influenced by consumer behavior patterns and external factors such as weather changes, which impact customer mobility. To address this problem, a web-based revenue prediction system was developed using the Random Forest Regression algorithm. This study integrates internal historical transaction data with secondary weather data (temperature and rainfall) from NASA POWER satellites. To overcome the inherent limitation of the Random Forest algorithm in recognizing temporal dimensions, a feature engineering technique was applied by creating time-lagged variables (lag features), referring to the Autoregressive Random Forest modeling approach.
The model performance evaluation using testing data showed a Mean Absolute Percentage Error (MAPE) of 11.17%, which is categorized as good forecasting. However, the obtained Coefficient of Determination R² was relatively low at 0.023. This low R² value indicates that while the model provides accurate nominal estimates on average, it has limitations in explaining extreme variability due to high daily noise. The findings suggest that extreme rainfall tends to act as a physical barrier to consumer mobility rather than an instant purchase trigger. Nevertheless, the findings indicate that the system is capable of providing daily revenue predictions based on historical data as a quantitative reference to support strategic decision-making at Apotek Mitra Medical.