Using ARIMA and Soft Computing Approaches to Predict the Volume of Consumer Complaints for Five Telecommunications Corporations in Taiwan

• Main Authors: Ting, Chih-Jung, 丁誌榮
• Other Authors: Shao, Yuehjen E.
• Format: Others
• Language: zh-TW
• Published: 2017
• Online Access: http://ndltd.ncl.edu.tw/handle/93500824731551636324

碩士 === 輔仁大學 === 統計資訊學系應用統計碩士在職專班 === 105 === A flourishing telecommunications industry in Taiwan contributes positively to the income of the Treasury Department. After the deregulation of the telecommunications industry in Taiwan, privately-run telecommunications companies have sprung up in the market. As a result of the fierce competition, customers have grown unhappy with the signal quality, fee calculation, and service quality of the telecommunications companies and gone on to file complaints. An increase in customer complaints can lead to a higher operating cost for telecommunications companies, for example, an increase in labor cost and change of operating strategies. Hence, a research undertaking to estimate the number of complaints and determine if the labor cost invested and operating strategies need to be modified can have very meaningful contributions to telecommunications companies. The current study used the monthly statistics of customer complaints provided by the National Communications Commission, Republic of China for each of the top five telecommunications companies in Taiwan (Chunghwa Telecom, Taiwan Mobile, FarEas Tone Telecommunications, Taiwan Star Telecom,and Asia Pacific Telecom) to create predictive models, employing the autoregressive integrated moving average (ARIMA) and soft computing methods. The soft computing methods used included artificial neural network (ANN), support vector regression (SVR), and multivariate adaptive regression splines (MARS). This study used mean absolute percentage error (MAPE) to compare the degree of accuracy of each predictive model. The study results showed that the predictive model based on ARIMA was more accurate than those based on ANN, SVR, and MARS.