顧客履約預測模式之使用與修正─應用類神經網路

Application of the Neural Network in the Usage and Revise of the Prediction Model on Customer’s Fulfill Appointment

劉立雯(Li-Wen Liu)

指導教授 ： 張百棧

元智大學/工程學院/工業工程與管理學系/碩士(2000年)

摘要

預約制度的優劣會影響企業之利潤與服務品質，而決定預約系統的成敗關鍵在於顧客履約率預測的準確性，以往曾有人使用類神經網路建立預測模式，並以顧客面為導向建立顧客屬性資料庫，主因類神經網路具備高速之處理速度，以及較佳之容錯能力，系統架構克成後，在使用操作及資料建立上均較容易，較能符合實際需要及服務業顧客需求多變且不穩定之特性。本論文即以該模式為基礎，針對其之缺失與不足之處加以探討與改善。
本論文的目的在建構一預測能力更佳之服務業顧客履約預測模式、建構一個可變動的顧客資料庫與修正預測結果中的異常值。本研究考慮兩項變動因子，一為前次失約紀錄，另一為修正因子 ，以台北榮民總醫院內科門診做為實證，針對預測結果中的異常值，利用修正因子加以修正。本研究的方法由於可隨時修改病患資料，比舊有模式更能應用於實際醫院預約狀況，且其準確率也相當高，達96%。修正因子法與傳統Re-training方法比較後也發現，傳統Retraining法之樣本準確率及整體準確率皆沒有修正因子法為佳。

關鍵字

類神經網路 ； 顧客履約預測模式 ； 變動因子 ； 前次失約紀錄 ； 修正因子 ； Re-training

並列摘要

The fit and unfit of the appointment system will do an effect on enterprise’s profit and the quality of service, and how to decide the crucial point of the appointment system’s quality is up to the accuracy of customer’s fulfill appointment prediction system. Recently, researches have applied the neural network to establish the prediction model on customer’s classification data. The neural network, which has a high speed of running and the high allowance of errors was easier to set up the data and to be used after the model established. It is more suitable in the real word application and also has the characters of satisfying customer’s variable needs and stabilizing the unsteadiness in service industries. So my thesis is based on the model, and aims at improving its shortcomings.
The purposes of my thesis are to establish a better accurate ratio of the prediction model on customer’s fulfill appointment , to set up a changeable customer’s database and to revise the error data of the results. My research considers two variable factors, one is the customer’s “Breaking on appointment last time” record, and the other is “ ”, the revise factor of the model. Taking the Internal Medicine clinic service of Taipei Veterans General Hospital for instance, to aim directly at the errors of the prediction model, to revise them by using the revise factor. The method can amend the customer’s data any time, and it is easier to be applied to the real prediction situation of hospitals. It also do a highly prediction ratio which is up to 96%. And after comparing my method with the traditional Re-training method, we find out which is the revise factor method is better than the individual sample prediction ratio and the whole samples’ ratio of the traditional Re-training method.

並列關鍵字

neural network ； prediction model on customer’s fulfill appointment ； variable factors ； Breaking on appointment last time ； revise factor ； Re-training

參考文獻

1. Chase R. B., “Where Does the Customer Fit in a Service Operation?,” Harvard Business Review, Nov-Dec 1978.

2. Fry J., “Appointment in General Practice.” Operational Research Quarterly, 15, 224~232, 1964.

3. Hertz P. and P. L. Stamps, “Appointment-keepings Behavior Re-evaluated.”, American Journal of Public Health, 67, 1033~1036, 1977.

4. Kimes S. E., “Yield Management: A Tool for Capacity-Constrained Service Firms.” Journal of Operations Management, Vol.8, No.4, 348~363, 1989.

5. Lovelock, C. H., “Classifying Services to Gain Strategic Marketing Insights,” Journal of Marketing, Vol.47, Summer 1983.