Comparison Various Analytical Approaches to Find The Most Efficient and Effective Method for Peak Hour Identification

Abstract

The Peak hour sales identification is essential to manage staff, inventory, and service capacity in coffee shop operations. This study compares an exploratory heatmap with two forecasting models, linear regression and Seasonal ARIMA (SARIMA) using six months of hourly transaction data from a coffee shop (1 March–17 August 2024) . The heatmap offers rapid visual recognition of high traffic periods but provides no predictive capability. For prediction, this study trained a linear regression and a SARIMA specification tuned by standard diagnostics; model performance was assessed on a held out set using MAE, RMSE, and MAPE. Linear regression yielded RMSE = 6.68, MAE = 5.40, and MAPE = 138.06%, indicating inadequate fit for intraday demand dynamics. In contrast, SARIMA achieved RMSE = 0.828, MAE = 0.557, and MAPE = 40.34%, substantially reducing error by explicitly modeling autocorrelation and recurrent seasonal cycles. The results show that seasonality aware time series modeling delivers actionable, interpretable forecasts for near term operational planning (such as staffing and product preparation). Overall, the proposed pipeline, heatmap for rapid situational awareness plus SARIMA for prediction, constitutes a practical baseline for peak hour identification in small scale retail.