TY - GEN
T1 - Uncovering hidden spatial patterns by hidden Markov model
AU - Huang, Ruihong
AU - Kennedy, Christina
N1 - Copyright:
Copyright 2021 Elsevier B.V., All rights reserved.
PY - 2008
Y1 - 2008
N2 - Many spatial data mining and spatial modeling approaches use Euclidean distance in modeling spatial dependence. Although meaningful and convenient, Euclidean distance has weaknesses. These include providing an over simplified representation of spatial dependence, being limited to certain spatial pattern and symmetrical relationships, being unable to account for cross-class dependencies, and unable to work with categorical especially multinomial data. This paper introduces Hidden Markov Model (HMM) as an attractive approach to uncovering hidden spatial patterns. The HMM assumes that a hidden state (factor or process) generates observable symbols (indicators). This doubly embedded stochastic approach uncovers hidden states based on observed symbol sequences using two integrated sets of probabilities, transition probability and emission probability. As an alternative to Euclidean distance based approaches, the HMM measures spatial dependency by transition probabilities and cross-class correlation better capturing geographic context. HMM works with data of any measurement scale and dimension. To demonstrate the method, we assume urban spatial structure as a hidden spatial factor underlying single family housing unit prices in Milwaukee, Wisconsin, we then use the HMM to uncover four hidden spatial states from home sale prices.
AB - Many spatial data mining and spatial modeling approaches use Euclidean distance in modeling spatial dependence. Although meaningful and convenient, Euclidean distance has weaknesses. These include providing an over simplified representation of spatial dependence, being limited to certain spatial pattern and symmetrical relationships, being unable to account for cross-class dependencies, and unable to work with categorical especially multinomial data. This paper introduces Hidden Markov Model (HMM) as an attractive approach to uncovering hidden spatial patterns. The HMM assumes that a hidden state (factor or process) generates observable symbols (indicators). This doubly embedded stochastic approach uncovers hidden states based on observed symbol sequences using two integrated sets of probabilities, transition probability and emission probability. As an alternative to Euclidean distance based approaches, the HMM measures spatial dependency by transition probabilities and cross-class correlation better capturing geographic context. HMM works with data of any measurement scale and dimension. To demonstrate the method, we assume urban spatial structure as a hidden spatial factor underlying single family housing unit prices in Milwaukee, Wisconsin, we then use the HMM to uncover four hidden spatial states from home sale prices.
KW - Data mining
KW - GIS
KW - Hidden Markov model
KW - Spatial modeling
UR - http://www.scopus.com/inward/record.url?scp=56749181990&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=56749181990&partnerID=8YFLogxK
U2 - 10.1007/978-3-540-87473-7_5
DO - 10.1007/978-3-540-87473-7_5
M3 - Conference contribution
AN - SCOPUS:56749181990
SN - 3540874720
SN - 9783540874720
T3 - Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
SP - 70
EP - 89
BT - Geographic Information Science - 5th International Conference, GIScience 2008, Proceedings
PB - Springer-Verlag
T2 - 5th International Conference on Geographic Information Science, GIScience 2008
Y2 - 23 September 2008 through 26 September 2008
ER -