Incorporating spatial association into statistical classifiers: local pattern-based prior tuning

ABSTRACT

This paper proposes a new classification method for spatial data by adjusting prior class probabilities according to local spatial patterns. First, the proposed method uses a classical statistical classifier to model training data. Second, the prior class probabilities are estimated according to the local spatial pattern and the classifier for each unseen object is adapted using the estimated prior probability. Finally, each unseen object is classified using its adapted classifier. Because the new method can be coupled with both generative and discriminant statistical classifiers, it performs generally more accurately than other methods for a variety of different spatial datasets. Experimental results show that this method has a lower prediction error than statistical classifiers that take no spatial information into account. Moreover, in the experiments, the new method also outperforms spatial auto-logistic regression and Markov random field-based methods when an appropriate estimate of local prior class distribution is used.

KEYWORDS:

• Spatial pattern
• statistical classifier
• spatial auto-logistic regression
• spatial data

Disclosure Statement

No potential conflict of interest was reported by the author(s).

Notes

1. $NCP>0$, $NCP=0$ and $NCP<0$ indicate positive, no and negative spatial associations, respectively. The larger the $NCP$ is, the stronger the spatial association.

2. The total number of comparisons for each dataset was one plus the triple of the number of categories in the dataset. The number of the comparison with true null hypothesis is used in the Bonferroni corrections.

Additional information

Funding

The work is supported by the Strategic Priority Research Program of the Chinese Academy of Science under Grant [XDA19040501], the Chinese National Science Fundation under Grant [41725006], the National Natural Science Foundation of China [Nos. 41871286, 61672331] and the Natural Science Foundation of Shanxi Province, China [No. 201701D121055].

Notes on contributors

Hexiang Bai

Hexiang Bai is an associate professor in Computer Science at the school of computer and informatin technology, Shanxi University. His research interests include rough set based granular computing and spatial statistics. E-mail: [email protected].

Feng Cao

Feng Cao is an associate professor in Computer Science at the school of computer and informatin technology, Shanxi University. His research interests include remote sensing and spatial statistics. E-mail: [email protected].

M. Peter Atkinson

M. Peter Atkinson is currently a professor of Spatial Data Science at Lancaster University and visiting professor at the Chinese Academy of Sciences, Beijing, China. The main focus of his research is in remote sensing, spatial data science and spatial and spatio-temporal statistics applied to a range of environmental, epidemiological, natural hazard and other spatial phenomena. E-mail: [email protected].

Qian Chen

Qian Chen is an associate professor in Computer Science at the school of computer and informatin technology, Shanxi University. His research interests include text mining and statistical machine learning. E-mail: [email protected].

Jinfeng Wang

Jinfeng Wang is a professor in Geograhphical Information Sciences at the Institute of Geograhpic Sciences and Natural Resources Research, Chinese Academy of Science. His research interest is spatial statistics. E-mail: [email protected].

Yong Ge

Yong Ge is a professor in Geograhphical Information Sciences at the Institute of Geograhpic Sciences and Natural Resources Research, Chinese Academy of Science. Her research interests include spatial statistics spatial scale transformation. E-mail: [email protected].