Using adaptive network-based fuzzy inference system to forecast automobile sales

Improving the sales forecasting accuracy has become a primary concern for automobile industry. Here, we only focus on new automobile sales in Taiwan. The data set is based on monthly sales, and the data can be divided into three styles of automobile sales. To address our concern, we developed a sales forecasting methodology that considers several variables such as current automobile sales quantity, coincident indicator, leading indicator, wholesale price index and income. First, we use the stepwise regression to select most influential variables as our input variables. Then, we input the influential variables and sales in adaptive network-based fuzzy inference system (ANFIS) to obtain the forecast. Finally, we compare our model with two forecasting models: autoregressive integrated moving average model (ARIMA) and artificial neural network (ANN). Empirical results demonstrate that the application of the ANFIS model outperforms the other two models. In addition, we modified the historical and holdout periods to improve forecasting accuracy while considering the impact from the financial tsunami in 2008.