An Efficient Incremental Spatial Association Mining to Explore Significant Rule Changes for Altered Spatial Databases

空間資料庫(Spatial Database)中儲存著空間資料(Spatial Data)與非空間資料(Non-spatial Data)，能提供空間資料型態(Spatial Data Type)與空間運算的功能，亦被廣泛應用在地理資訊系統(GIS)、遠端遙測系統(Remote Sensing)、醫療影像、都市規劃等各種領域上。目前許多空間資料探勘(Spatial Data Mining)的研究都是衍生自應用在交易型資料庫(Transactional Databases)上的方法，但這些方法僅能應用在非重複(Non Recurrent)物件上，然而一張圖片中出現大量且重複(Recurrent) 的空間物件是普遍且常見的情形。其次，隨著時間增加，空間資料庫中的資料也會與日俱增，因此先前所探勘出來的空間關聯規則(Spatial Association Rules)可能無法反映異動後空間資料庫中的空間型樣(Spatial Pattern)。目前甚少有學者在研究空間關聯規則的改變(Spatial Association Rule Change)。如能夠找出空間關聯規則更新前後的差異，便可將這些資訊應用於各個領域中，如生態學、都市規劃等等，提供給決策者更豐富的資訊。再者，空間資料探勘的目的之一是產生含空間距離(Spatial Distinct)的空間關聯規則，如能在產生空間關聯規則的過程中，使用模糊理論(Fuzzy Theory)，將距離語義化(Semantic)，將可產生較貼近使用者認知的空間關聯規則。基於上述理念，本論文進行下列四件工作之研究：(1)對被探勘的空間資料進行預前處理(Preprocessing)，以完整地呈現空間知識(Spatial Knowledge)，提供決策者進行決策參考；(2)增加漸進式資料探勘的概念，提出高頻空間型樣Frequent Spatial Pattern Growth之漸進式資料探勘的演算法。此方法不需重新掃瞄整個資料庫，僅針對新增的資料進行探勘，以更新高頻空間型樣，因而能提升探勘更新後空間關聯規則之效能；(3)建立相似度指標 (Similarity Measure)及相異度指標 (Difference Measure)來判斷在兩個不同時期間所探勘出的空間關聯規則有哪些規則發生變化及其改變程度；(4)加入了模糊理論的概念，將明確的距離值語義化，讓探勘出的空間關聯規則更貼切使用者的認知。

關鍵字

空間資料庫；空間關聯規則；漸進式資料探勘；規則改變；模糊理論

並列摘要

So far, many researches on spatial data mining derive from approaches that are originally used in transactional databases. Moreover, those methods are only practicable with non-recurrent objects. However, data in spatial database increase with time and images with a great quantity of recurrent identical spatial objects are more useful and realistic. Therefore, we propose an efficient and effective incremental spatial association mining method for addressing the problem that the infeasible mining methods can not be applied to the spatial databases well. There are three phases in the proposed method. Phase one is the generation of recurrent spatial patterns. We also apply fuzzy theory to refine the spatial association rules such that these rule sets can include semantic. An incremental mining method, FSP-split (Frequent Spatial Pattern-split) is proposed for exploring the frequent spatial patterns in an altered spatial database. This method is fast because it doesn’t require another scan over the whole database while database is updated. The third phase is to detect the changes between spatial association rule sets in two time periods and calculate the degree of change. Further, our method can calculate the degrees of change to discover the significantly changed rules. The variations of the spatial rule changes over time can be provided for decision-making. Experiments are implemented in both transaction databases and spatial databases. The experimental results in transaction databases demonstrate that our proposed incremental mining method, FSP-split, performs efficiently. The experimental results in spatial databases show that the similarity and unexpectedness measures can detect the rule changes at the difference time.

並列關鍵字

Spatial Association Rules ； Incremental Data Mining ； Rule Changes

國際替代計量

An Efficient Incremental Spatial Association Mining to Explore Significant Rule Changes for Altered Spatial Databases

全文下載

主題瀏覽