多元交通行動服務（Mobility as a Service, MaaS）為近年來一項新興概念，國內外許多城市已開始推動MaaS，了解民眾的套票購買行為及考量因素將有助於主管機關與業者研擬適當的行銷策略以推廣MaaS。本研究以資料探勘方法建立會員方案續買預測模型，並以高雄市MaaS系統會員註冊資料、方案購買記錄及電子票證搭乘記錄為輸入資料。為能改善MaaS會員套票購買資料在各方案間之資料不平衡問題，本研究提出以機率分配為基礎之增加抽樣方法（probability distribution-based over-sampling, PDB），並將此方法與文獻中常用的SMOTE（synthetic minority over-sampling technique）方法以網路上公開的資料集進行測試與比較，結果發現PDB顯著優於SMOTE，可進一步應用於處理MaaS會員資料，並建構MaaS會員方案續買預測模型。經由交叉驗證測試結果發現，經由PDB增加抽樣之資料所建構的決策樹及支持向量機模式皆有不錯預測結果，顯示模型具備會員方案續買預測能力。此外，本研究也針對決策樹模型的分支規則進行探討，發現除使用者在各個運具的每月花費會影響續買行為外，月份也是重要的考量因素。

Mobility as a service (MaaS) is an emerging concept in recent years, and it has been promoted in many cities around the world. Understanding users' purchase behavior of MaaS packages will help the authorities and operators to promote MaaS. This study develops a package-purchasing prediction model using data mining techniques. The prediction model is built and trained using the data of membership registration, package-purchasing records and iPASS card transit-ridership records of MaaS users in Kaohsiung. Through preliminary data processing and analysis, it is found that significant imbalance exists in MaaS users' package-purchasing records for the different plans, and the imbalance may affect the prediction results of the model. To address this issue, the study proposes an oversampling method, namely probability distribution-based over-sampling (PDB), to generate additional samples. This method is first tested and compared with the SMOTE (synthetic minority over-sampling technique) method commonly used in the literature by using datasets published online, and it is found that the proposed method is significantly better than the SMOTE method. Then the study uses the method to balance the MaaS users' package-purchasing data, and constructs a decision tree model and a support vector machine model for MaaS users' package-purchasing prediction. Through cross-validation test results, it is found that the models constructed by the oversampling data using the simulation method has good prediction results which shows that the prediction model has the ability to predict the users' purchase behavior. This study also discusses the branch variables in the decision tree model, and found that the user's monthly spending on each public transportation mode will affect the package-purchasing of MaaS users. Moreover, month is also an important variable. The results can be used as a reference for MaaS operators to take appropriate actions on marketing based on the results of the users' package-purchasing prediction.