以啟發式區域劃分演算法求解地價區段劃分問題

地價區段劃分為土地估價作業中最重要的環節之一，但現階段仍需要依照人工主觀判定，對於土地估價之公平性和公正性皆有影響。本研究將地價區段視為p-region問題，並提出一種啟發式演算策略。先以地價區段劃分之現況為基礎，利用先分群再求解的啟發式演算策略，結合不同區域劃分演算法，再配合變異係數作為閾値改善地價區段劃分，最後導入空間與非空間的區域劃分評估指標來針對改善的成果進行比較分析。研究結果顯示，提出的演算策略在處理大量土地資料更有效率外，同時改善現有地價劃分結果，以及提升區段內與區段間地價的同質性。此外，所使用的五種區域劃分演算法中，不論在計算時間或劃分結果方面，region-k-mean具有最佳的平衡表現。再者，本研究也發現邊界輪廓係數與演算法改善之結果具有一致性，相較於其他非空間概念的指標更適合應用在區域劃分之評估。

關鍵字

區域劃分；空間鄰接性；變異係數；地理輪廓係數

並列摘要

The assessment and demarcation of land value sections is a crucial aspect of current land appraisal operations; however, these tasks often rely on human experience and can leave the quality of the results uncertain to the public. This study aims to design and implement a heuristic procedure to deal with a large number of land parcels and to demarcate land value sections. The study achieved this goal by using the divide-and-conquer paradigm to filter current sections using a threshold of coefficient of variations (cv), solving the land demarcation as a p-regions problem using various regionalization methods, and comparing the results based on both spatial and non-spatial evaluation criteria. The results showed that the proposed strategy can improve the within-section homogeneity and inter-section heterogeneity, while considering spatial integrity as well. Among the five algorithms used, regional-k-means was found to have the best balance of performance in terms of computing time and regionalization results. Additionally, the boundary silhouette was found to be a more stable index for evaluating the quality of regionalization when compared to other non-spatial criteria used in the study.