A Support Vector Based Hybrid Forecasting Model for Chaotic Time Series: Spare Part Consumption Prediction

Reliability of spare parts inventory in the company is one of the most significant challenges in the field of maintenance and repairs, but on the other hand, the liquidity crisis resulting from the purchase of surplus spare parts is another challenge facing the organizational financial field. Accordingly, accurate forecasting of future consumption is one of the most important solutions for inventory control systems. But because of the impact of so many variables on spare part consumption, most real-world data is chaotic. This leads to the use of classical methods to predict future demand, with high error and low reliability. In this research, a novel and reliable hybrid model based on the support vector machine (SVM), and two single algorithms (STL Decomposed ARIMA and three-layer feed-forward neural network) has been presented to predict the future consumption of spare parts. The proposed model (SVM-ARIMA-3LFFNN hybrid model) also experiments on several chaotic time series in the rapid miner repositories. The forecasting results indicate that the proposed hybrid model attains superior performance compared with a single model and can adapt to chaotic time series. Performance criteria considered in this study are MAE, RMSE, MAPE, and sMAPE. The results indicate that the proposed model can improve the RMSE, MAPE, and sMAPE (up to 30% improvement).