應用資料探勘技術於預測潛在客戶之研究-以臍帶血公司為例

NCU Institutional Repository > 管理學院 > 資訊管理學系碩士在職專班 > 博碩士論文 > Item 987654321/72127

請使用永久網址來引用或連結此文件: http://ir.lib.ncu.edu.tw/handle/987654321/72127

題名:	應用資料探勘技術於預測潛在客戶之研究-以臍帶血公司為例
作者:	黃柏中;Chung,Huang-Po
貢獻者:	資訊管理學系在職專班
關鍵詞:	資料探勘;監督式學習技術;預測模型;單一分類技術與多重分類技術;data mining;supervised machine Learning techniques;prediction model;single classification technique and multiple classification technique
日期:	2016-06-07
上傳時間:	2016-10-13 14:27:21 (UTC+8)
出版者:	國立中央大學
摘要:	我國新生兒出生率每年持續下降的因素，加上臍帶血、臍帶幹細胞之治療上受限於醫療法規的限制，使得該產業的競爭程度相當激烈。而在如此競爭激烈的產業之中，如何有效的運用潛在客戶名單資料，使潛在客戶成為幹細胞存戶，便是幹細胞儲存業者在營運上的重要課題。本研究著重在潛在客戶之落點分析，依現有所蒐集到的潛在客戶名單資料，利用資料探勘監督式學習技術所分析出來的資訊，提供個案公司決策者或執行的業務單位有所依據。以往傳統的經驗法則在執行上可能因缺乏客觀的信度與效度，導致目標客戶落點不準確，而喪失企業獲利的機會。本研究的目的包括利用資料探勘監督式學習技術挖掘出潛在客戶的客戶落點預測模型、建立潛在客戶落點預測模型，並進一步比較單一分類技術與多重分類技術之差異及藉由本次研究，提供個案公司在預測潛在客戶的相關產業做為參考。本研究在實驗流程上採用Weka資料探勘軟體，並進行不同分類技術的實驗，本研究在單一分類技術分別採用決策樹推估模式、支援向量機推估模式、類神經網路推估模式、最鄰近演算法等四種單一分類技術，並搭配多重分類技術中的Bagging、AdaBoost以加以驗證，以試圖獲得最佳潛在客戶落點預測模型。經過實驗結果得知以2014年的訓練集資料而言，在單一分類技術中以最鄰近演算法表現最佳，在多重分類技術中分別以Bagging的類神經網路推估模式、AdaBoost的決策樹推估模式表現最佳，透過Weka的實驗結果，正確率(Correctly Classified Instances)與接收者操作特徵曲線(ROC)普遍值達到0.68、0.7左右，具有較佳參考意義。因此，本研究建議個案公司未來在進行潛在客戶落點預測時，可以優先採用單一分類技術中的最鄰近演算法，並搭配多重分類技術中Bagging的類神經網路推估模式、AdaBoost的決策樹推估模式，以進行潛在客戶落點預測分析。 ;Since the birth rate of new born babies in our country was continued to decline every year and the applications of treatments with cord blood and umbilical cord stem cells were restricted by medical regulations, the competitions in the industry were extremely fierce. In such competitive industry, how to use the name list of potential customers effectively and making the potential customers become existing stem cell storage customers was an important business operation issue of the stem cell storage providers. This study focuses on the analysis of potential customer placements. Based on the collected data name list of potential customers and using the result information analyzed with supervised machine learning techniques for data mining, we could provide the corporate decision makers or the executive business unit useful reference information. Due to lack of objective reliability and validity, using traditional rule of thumb resulted inaccurate placement of target customers and loss of company profit opportunities. The aims of this study were discovering the best prediction model for potential customer placements by supervised machine learning techniques; further comparing the difference between single and multiple classification techniques; and with this study, providing case company reference information for prediction of potential customers in related industries. For trying to get the best prediction model of potential customer placements, the experimental processes was designed to use Weka data mining software and compared different classifying techniques. In this study, we adopted four kinds of single classification techniques, which are decision tree, support vector machine, artificial neural network, and k-nearest neighbor. Besides, the Bagging and AdaBoost methods are employed to construct classifier ensembles of the four single classifiers. The experimental results show that with the training data of 2014, the nearest neighbor classifier provides the best performance. For classifier ensemble, the Bagging based artificial neural network and the AdaBoost based decision tree models perform the best. Particularly, the classification accuracy and receiver operating characteristic curve (ROC) can achieve about 0.68 and 0.7. Therefore, we could suggest the case company to adopt the nearest neighbor algorithm first to perform the prediction of potential customer placements and use both Bagging based neural network and the AdaBoost based decision tree models to perform potential customer placement prediction at the same time.
顯示於類別:	[資訊管理學系碩士在職專班 ] 博碩士論文

文件中的檔案:

檔案	描述	大小	格式	瀏覽次數
index.html		0Kb	HTML	485	檢視/開啟

在NCUIR中所有的資料項目都受到原著作權保護.

社群 sharing

資料載入中.....