資料探勘分類技術於游泳會員流失區別模型之研究

A Study of Swimming Member Churn Model Using Data Mining Classification Techniques

本研究旨在暸解台灣師大本部游泳會員之組成結構，運用資料探勘中的鑑別分析、類神經網路、多元適應性雲形迴歸以及整合類神經網路與多元適應性雲形迴歸等分類技術建構台灣師大游泳會員流失區別模型，並瞭解會員流失的重要特徵。台灣師大本部游泳會員資料經整理後，共2,707筆，在剔除內容不合之資料後，共計2,380筆。本研究結果如下：一、會員組成結構如下：（一）男（49.87%）、女（50.13%）會員幾近相等；（二）會員類型以自由會員（55.63%）稍多；（三）大部分會員居住在大安中正兩區（79.12%）；（四）會員平均年齡為32.15歲且年齡的分佈平均沒有特別集中的現象；（五）會員平均會齡為0.69年且大量集中在二年以下（93.66%）；（六）沒有折扣會員（95.80%）佔大多數；（七）繳費金額以4,500元（42.86%）最多；（八）購買季節以夏季（49.66%）居多；（九）會員使用時段以不受限制的任何時段（56.64%）最多。二、整合類神經網路與多元適應性雲形迴歸分析模型的整體分類績效最高，為84.03%。整合模式成功地建構台灣師大本部游泳會員流失區別模型。三、台灣師大本部游泳會員流失的重要特徵為會齡在1年以下、繳費金額為2,500元、購買季節在夏季的一般類型會員。有鑑於此，台灣師大游泳池管理單位可就分析的結果轉為會員維繫方案，達到降低會員流失的目的。

The purpose of this study was to construct a model for analyzing the turnover rate among those who become members of the NTNU main campus swimming pool, in order to facilitate the diminishing, on the part of the swimming pool administrative staff, of member dropout and turnover. This model was constructed by using data mining classification technology including discrimination analysis, artificial neural networks, multivariate adaptive regression splines, and multivariate adaptive regression splines. The first step was of course to establish the salient characteristics of swimming pool member turnover. After reorganizing the data for all swimmers in the NTNU main campus swimming pool, 2,707 records were chosen as our initial data. After deleting the unreasonable data, a total of 2,380 records were discussed in this study. The research results were as follows: 1. As for the over-all constitutive structure of this group:（1） the number of male members（49.87%） and female members（50.13%） was almost equal;（2） 55.63% were "free" members directly connected to NTNU;（3） most members were residents of Da-an and Zhongzheng districts（79.12%）;（4） the average age of members was 32.15 years, and there was a roughly even age distribution;（5） the average participation period of members was 0.69 years and most joined for less than 2 years（93.66%）;（6） the vast majority were non-discount members（95.80%）;（7） the most frequently levied membership fee was 4,500 NT dollars（42.86%）;（8） the summer season was easily the most popular for enrollment（49.66%）;（9） no limitation on times when the pool could be used was the most popular choice（56.64%）. 2. Combining artificial neural networks and multivariate adaptive regression splines generated a classification rate of 84.03%. The integrated approach successfully constructed a member churn model for the NTNU main campus swimming pool. 3. The optimal characteristics of an NTNU main campus swimming pool member were: participation period below 1 year, member fee of 2,500 NT dollars, membership enrollment in summer, and general members. It is hoped that the NTNU Swimming Pool Administration Department can use these results in order to decrease member dropout and turnover.