A comparative study of demand forecasting based on machine learning methods with time series approach

Demand forecasting can have a significant impact on reducing and controlling companies' costs, as well as increasing their productivity and competitiveness. But to achieve this, accuracy in demand forecasting is very important. On this point, in the present study, an attempt has been made to analyze the time series related to the demand for a type of women's luxury handbag based on a framework and using machine learning methods. For this purpose, five machine learning models including Adaptive Neuro-Fuzzy Inference System (ANFIS), Multi-Layer Perceptron Neural Network (MLPNN), Radial Basis Function Neural Network (RBFNN), Discrete Wavelet Transform-Neural Networks (DWTNN), and Group Model of Data Handling (GMDH) were used. The comparison of the models was also based on the accuracy of the forecasting according to the values of forecasting errors. The RMSE, MAE error measures as well as the R, correlation coefficient were used to assess the forecasting accuracy of the models. The RBFNN model had the best performance among the studied models with the minimum error values and the highest correlation value between the observed values and the outputs of the model. But in general, by comparing the error values with the data range, it is concluded that the models performed reasonably well.