Improving Accuracy of Time Series Forecasting by Applying an ARIMA-ANN Hybrid Model

Accurate demand forecasting is critical for any small and medium-sized manufacturer. Limited structured data sources commonly prevent small and medium-sized manufacturers from improving forecasting accuracy, affecting overall performance. We classified products, then implemented a hybrid forecasting method and compared the outcome with Exponential smoothing, ARIMA, LSTM, and ANN forecasting techniques. Numerical results demonstrate that a selection of forecasting methods is not independent of product categorization. For slow-moving products, careful consideration is required. The hybrid ARIMA-ANN method can outperform some existing techniques and lead to higher prediction accuracy, by capturing both linear and nonlinear variations.