本研究希望能在預測離職模型上貢獻一份心力，藉由數據分析方法中的決策樹演算法建立預測離職模型。利用此種演算法是因兩大優點：一個是分析結果圖像化，不需要太多的數學基礎就能看懂且運用；另一個就是其計算速度快，在資料量越大的時候越能凸顯此項優點。擁有這些優勢讓他在實務上的運用具有高潛力，也有機會能更加快速的反應組織成員的狀況。 本研究梳理了眾多文獻中能用來預測離職行為的因素，參考並作為輸入決策樹預測變數的基礎，輸出則是離職狀態與否。我們就(在職/離職)者在過去的同一時點出發，到了離職者正式離職，而在職者依然在職的期間，他們的狀態改變中，顯著變項作為預測行為的因素，並試圖探討此種因素的形成可能。 這次的資料收集中，我們從個案公司上取得一個高科技部門專案研發團隊的EHRMs資料，以五年為一單位，共237位員工、889 筆(因其不是每個員工年資都在5年以上，故資料總數不為237的5倍)人。我們建立預測離職模型的因素有二：年資、請假次數總計 此二項分別再組成了六組結果： (1)請假次數多於43次者，幾乎離職 (2)請假次數小於(含)43次者，同時年資大於48個月，幾乎在職 (3)年資小於(含)48個月者，同時請假次數大於18次，多離職 (4)請假次數小於(含)18次者，同時年資大於27個月，幾乎在職 (5)年資介於27(含)~3個月間者，請假次數小於(含)18次者，較多留職 (6)年資小於(含)3個月者，請假次數小於(含)18次者，較多離職 ;Base on the approach of machine learning, we built an employee turnover prediction model faster and more visualization by utilizing Decision tree on human resource management (HRM) data. These advantages allow it be more useful in practice and reflect the condition of labor force in firm timely.
In this thesis, we refer to many factors of predicting turnover behavior from literatures, and collect these factors from EHRMs of the firm. We find out some significant changes of those who left or stayed within five years. Then, we build a turnover prediction model by these changes.
We totally get 889 records of 237 employees from the firm. The result of the model is : (1) Those who took leaves more than 43 times, 95% of them leave (2) Those who took leaves less than or equal to 43 times and had more than 48 months’ seniority, 0.2% of them leave. (3) Those who took leaves between 18 to 43 times and had less than or equal to 48 months’ seniority, 69% of them leave. (4) Those who took leaves less than or equal to 18 times and had between 27 to 48 months’ seniority, 5.3% of them leave. (5) Those who took leaves less than or equal to 18 times and had between 3 to 27 months’ seniority, 33% of them leave. (6) Those who Those who took leaves less than or equal to 18 times and had less than or equal to 3 months’ seniority, 64.2% of them leave.
