網際網路技術成熟及行動網路速度提升，使得電信業者提供之語音服務被網路服務取代。以營收貢獻來說，企業客戶遠大於一般消費客戶，且若企業客戶部分業務一旦流失至競業，其他相關業務皆可能轉至競業，期能保留目前使用語音服務的客戶，維持現有客戶之語音服務營收。本研究主題為企業客戶語音流失預測，為找出可能流失客戶，運用機器學習之方法建立預測模型，並考量企業客戶特性，納入企業客戶專有變數，以提升預測準確率。根據真實的電信客戶資料的研究結果顯示，企業客戶專有變數在前十大重要變數裡佔了三個；此外，依據預測結果對可能流失之客戶進行挽留，透過設定最適合閾值後，使營收損失最小化，相較於單純不作為，則可有效降低營收損失。

Due to the maturity of IP network technology and the provision of 4G mobile high speed networks, Internet-based services have become more popular. Nevertheless, the revenue of voice services for telecom operators has been substantially reduced. Yet the construction cost of broadband network and mobile phone base station remain the same. As a result, the profit of telecom operators has been drastically reduced. In addition, reports from the NCC shows that Taiwan’s telecommunications market has been saturated. Therefore, customer retention and customer churn management become important issues for telecom operators. In this work, we engage in the study of predicting enterprise customer churns in telecommunication industry because enterprise customers contribute more revenues to telecom service providers. Various variables, including the enterprise customers’ unique variables, have been identified, and the Xgboost algorithm is used to establish the prediction model. Our experimental results based on the real telecom customer data show that three enterprise customers’ unique variables are among the top 10 most important variables. In addition, our proposed prediction model is able to increase AUC and recall rate by 3% and 5.4% respectively, when compared to the prediction model that simply incorporates variables identified by previous work. We further try to minimize revenue loss by setting the most appropriate threshold. The experimental results show that by setting the threshold at 0.72 and applying customer retention strategy to the predicted customers, we are able to reduce the revenue loss by 525 units per customer.