| Summary: | 碩士 === 國立中興大學 === 企業管理學系所 === 107 === The purpose of this thesis is to develop a predictive model that won’t be too complicated for people to use and can be more accurate to predict the demand for the time series data of Taiwan''s machine tools from January 1998 to December 2018.
The prediction methods used in this paper include time series, neural network (back propagation network), and SARIMABPN hybrid model that mixes the two methods that mentioned before. In addition, variable industrial production index, real effective exchange rate index, and stock index have been included to see will the model be more accurate.
R language program has been use to analyze and predict this data set. After dividing the data into training set and test set. Using training set at first to choose the parameters combinations that may be most suitable for this data by adjusting the parameters such as learning rate, threshold, stepmax to get the most suitable result for each model. And then detect if the parameters combinations we choose through training set are still precise in test set.
And then compare which model is the best to this data by using the criteria for comparing the accuracy of various models like average absolute percentage error (Mean Absolute Error; MAE), Mean Square Error (MSE), and Average Absolute Error (MAE).
After comparison, found that the SARIMABPN hybrid model with considering industrial production index, real effective exchange rate index, and stock index variable gets the best results no matter on MAPE, MSE or MAE of the eight models which are built in this paper.
|
|---|
