Applying Artificial Neural Network and Decision Tree to Going Concern Doubt Prediction

• Main Authors: LO, YEN-HSIANG, 羅彥翔
• Other Authors: CHI, DER-JANG
• Format: Others
• Language: zh-TW
• Published: 2019
• Online Access: http://ndltd.ncl.edu.tw/handle/95fezp

碩士 === 中國文化大學 === 會計學系 === 107 === In recent years, all domestic industries have faced major challenges in industrial transformation, such as business model, cost adjustment and even service model. If the company is slightly inadvertent or inaccurate in decision-making, it will suffer a major loss, which will cause a major impact on the company's physique. As a result, the operator itself or the vast number of investors will suffer significant losses. Although there are many studies related to continuing business doubts, in view of this research, we will use China's electronics industry and traditional industries as research samples to establish a forecasting model for continuing business doubts, hoping to check or invest in investors and related practitioners. It is helpful to make decisions. The research model proposed by this research is divided into two stages. Firstly, the important variables are selected by using the neural network-based routing variable group. Then, the variables are respectively put into C5.0 and CHAID in the decision tree to establish a classification prediction mode. Through the five cross-validation procedures of this study, it is found that the traditional business forecasting model based on ANN+C5.0 has the best performance, the average predicted hit rate is 95.08%, and the type II error rate is 17.259%. In the electronics industry, the ANN+CHAID model has the best predicted average hit rate of 90.70%, and the average type II error rate is 25.119%. In addition, Tobin's Q, ROE, stock price ratio and free cash flow are repeated in the forecasting models of the two industries in the key indicators identified in this study. It is recommended that practitioners can refer to these indicators, plan check operations and conduct related Evaluation.