Spatially Weighted Bayesian Classification of Spatio-Temporal Areal Data Based on Gaussian-Hidden Markov Models

This article is concerned with an original approach to generative classification of spatiotemporal areal (or lattice) data based on implementation spatial weighting Hidden Markov Models (HMMs). In the framework this model at each unit specified by conditionally independent Gaussian observations and first-order chain for labels call it local HMM. The proposed modification conventional HMM spatially weighted estimators HMMs parameters. We focus rules Bayes discriminant function (BDF) plugged in parameter obtained from labeled training sample. For HMM, regression coefficients variances two types transition probabilities are used levels (higher lower) weighting. average accuracy rate (ACC) balanced (BAC), computed confusion matrices evaluated a test sample, as performance measures classifiers. methodology illustrated simulated real dataset, i.e., annual death collected Institute Hygiene Republic Lithuania 60 municipalities period 2001 2019. Critical comparison classifiers done. experimental results showed that higher level majority cases have advantage spatial–temporal consistency over one lower