點資料誤差對於空間型態分析之影響

Abstract

空間資料由於具有同時表示位置以及屬性的特點，而點資料是空間資料中最能精確表達位置的資料類型，因此在各領域中所扮演的角色日益重要。然而也因為點資料能夠精確表達位置之資訊，導致了點資料在傳達、散布時極易造成個人隱私之洩漏。為避免對隱私權造成危害，權責單位在發布資訊時多以行政區為單元，將點資料以加總後之形式公開。但以加總資料進行發布之方式，易受加總單元之空間尺度和研究範圍影響，進而使得空間分佈型態產生扭曲。 本研究試圖以對個別點資料置入隨機誤差之方式，以達到在維持空間分佈型態下仍可以保護隱私之目的。為檢定是否確實達到隱私保護之目的，採用平均最鄰近距離作為隱私保護之最低標準，並以蒙地卡羅方法模擬對點資料置入隨機誤差後，其對空間型態分佈造成之影響，以瞭解置入隨機誤差後空間型態是否產生扭曲之現象。 研究中依照點資料分佈之密度，將研究區域分為山區、鄉區、城區。以大於等於兩倍標準差與否，判定點資料之熱源區域，採用type1 error（原始資料中為熱源區域，添加隨機誤差後遭判定為非熱源區域）、type2 error（原始資料中並非熱源區域，添加隨機誤差後遭判定為熱源區域）作為比較之基準分別檢視其空間型態的變化。 研究結果顯示城區、山區分別在type1 error、type2 error之增加速度最快，鄉區則不論type1 error、type2 error都呈現緩慢的成長。在同時考慮保護隱私以及維持空間型態分佈之前提下，城區、鄉區有較好之表現，山區則相對較差。但總體而言，本研究之方法不失為一個能有效保護隱私且維持空間分佈型態的方式。惟隨機誤差之參數設定，仍必須視研究區域之個別特性以及決策者對於隱私保護之要求重新判斷。

Spatial data plays a crucial role in many research fields because of its ability of expressing multiple types of information such as location and other attributes at the same time. Point data is one major type of spatial data that is capable of demonstrating accurate and precise locations. However, this could probably contribute to a serious problem in privacy and security. Consequently, the government usually adopts aggregated data to release public information for privacy protection. Nevertheless, under the perspective of academic research, aggregated data is somehow easily influenced by spatial scale and the extent. This study aims to figure out the best measurement for point data spatial analysis, meanwhile protecting the privacy from releasing unexpected information. Random error introduction is proposed in this study to improve the spatial pattern distortion and privacy protection. Average nearest neighbor distance was applied as the threshold for privacy leak. Monte-Carlo simulation was used for simulating the spatial pattern distortion after introducing random error to the original point data. According to the spatial distribution of the original point data, the study area was separated in to three areas: mountain, county, and city. The result shows density of city and mountain areas grow fastest with type 1 and type 2 error; on the other hand, the country area grows slower with both type 1 and type 2 error. For the purpose of maintaining the spatial pattern and protecting privacy at the same time, the simulation result of city and mountain areas are relatively better than that of county area. In the conclusion, random error introduction can sucessfully protect privacy and meanwhile keep the spatial pattern effectively. However, the effect depends on the property of individual area.