Extension of eigenvectir spatial filtering approaches for large and diverse spatiotemporal datasets

2019 Fiscal Year Final Research Report

• Project/Area Number: 17K12974
• Research Category: Grant-in-Aid for Young Scientists (B)
• Allocation Type: Multi-year Fund
• Research Field: Geography
• Research Institution: The Institute of Statistical Mathematics
• Principal Investigator: Murakami Daisuke 統計数理研究所, データ科学研究系, 助教 (20738249)
• Project Period (FY): 2017-04-01 – 2020-03-31
• Keywords: 空間回帰

Outline of Final Research Achievements

This study develops fast and flexible spatial (and spatiotemporal) regression approaches for large and diverse geo-spatial datasets. This development is done by extending the random effects eigenvector spatial filtering (RE-ESF) approach, which is a spatial regression approach. First, we improve computational efficiency of RE-ESF by incorporating a pre-conditioning algorithm. Then, the memory consumption is drastically reduced for modeling very large samples through parallelization. After that, the developed fast approach is extended for spatio-temporal data, hierarchical data, non-Gaussian data, and other data by introducing latent variables capturing data properties. Usefulness of the proposed approach is verified by applying it to a wide variety of spatial and spatiotemporal data modeling. Finally, all the developed methods are implemented an R package spmoran to make them available for public.

Free Research Field

空間統計

Academic Significance and Societal Importance of the Research Achievements

近年急増する大規模な地理空間データを柔軟に解析するための空間回帰法を幅広く開発した。空間回帰法は空間疫学、空間計量経済学、計量地理学といった関連分野の高度化を、大規模データの解析手法の高度化の観点から後押ししうるものである。開発した各手法は既に統計ソフトウェアRのパッケージ化して一般公開済みであり、既に幅広い関連研究者に利用されている（例えば2019年度は7684回ダウンロードされた）。以上に加え、本研究で提案した空間回帰法は計算時間とメモリ消費が極めて小さく学術的にも新規的である。