Post office location analysis using geographic information system and data mining techniques

近年來由於政府實施無紙化及金融業者推行電子帳單的成效卓越，使得國內郵件的收寄量逐年下滑，郵局如何與民營業者競爭國內物流市場並達成盈餘目標，成為營運中不可忽視之因素。 傳統的郵局設點多依據公司規定與配合政府政策需求，甚少採用涉及複雜因素之區位分析進行選址。因此，如何有效且公正地評選郵局新設據點以提高收益，成為亟待解決之問題。 本研究目的在於提供高收益之郵局設點建議，我們提出一種評估中華郵政公司設點效益的方法，以國內郵局實際設點位置與相關空間資料來建置實驗模型。研究結果顯示，以本研究方法建立之預測模型可成功的提供中華郵政公司建議於何處新增據點可收最大功效。 我們首先蒐集中華郵政公司設點之鄰近區域資料，如競爭者設點數、人口因素、重要交通路口、郵件收寄量等。其次導入資料探勘技術分析影響郵件收寄量之因素，建立中華郵政公司設點收寄量預測模型。然後依照建立預測模型時所得到之區辨力分數，判斷採用何種資料探勘技術建立預測模型較適當。最後將所選定的預測模型套用於台北縣市各村里建物重心，透過環域資料分析以計算預估之收寄量，再整合各資料探勘技術之預測結果後推論出最佳設點建議。 實作中，以台北縣市資料來測試我們的方法。實驗數據顯示，我們的方法成功地找出十一個建議設點的村里，可提供給中華郵政公司作為高收益的設點建議。The amount of postal mail declines in recent years due to the efforts of paper-reduce policies implemented by the government, the industries, and the general publics. It becomes one of the important issues of the Chunghwa Post Company, to compete with other companies in domestic freight and mail services and to achieve the desired profits. Traditionally, the location of post offices were decided according to the government policies as well as the company regulations. The issues involved in the site selection analysis were seldom considered. Hence, developing an effective and fair mechanism to find the new post office locations that could improve the company’s surplus becomes an important problem to be solved. The purpose of this thesis is to provide recommendations to the post office site selection which will yield high profit to the company. We proposed a method to evaluate the effective profits that could be produced by a particular post office through the data mining techniques and the related GIS information. We first collect various data, such as neighborhood population, traffic flow, postal mail received at particular post office, competitor’s information, etc., and analyze these data using data mining techniques in order to establish prediction models. The most appropriate model was chosen to find the new post office sites. The Metropolitan Taipei area was chosen to illustrate our idea. The best sites for new post offices were selected through the buffering analysis as well as the data mining techniques. The experimental results show that our method can successfully find eleven locations which could generate most profit to Chunghwa Post Company if the new post offices were located in these places