Journal of Spatial Information Science | |
The role of location and time among other features when training a machine | |
Mahdi Hashemi | |
关键词: Inductive learning; Spatial data; Temporal data; Spatial autocorrelation; Temporal autocorrelation.; | |
学科分类:计算机科学(综合) | |
来源: University of Maine | |
【 摘 要 】
This study poses the question, how the recorded location and time for training samples should contribute in training and testing a machine? Historically, location and time have been either (a) dismissed altogether, (b) considered as the only input features, or (c) considered as additional input features along with other non-spatial and non-temporal features, when training or testing machine learning techniques. While experiments with real data reveal different generalization accuracies and prioritize these strategies in different ways, this paper argues that all these approaches fail to capture effectively how the spatial-temporal phenomena behave in reality. To that end, new theoretical methods are required to quantify the expert knowledge of spatial-temporal data peculiarities and effectively exploit them in machine learning techniques.
【 授权许可】
CC BY
【 预 览 】
Files | Size | Format | View |
---|---|---|---|
RO201902188270494ZK.pdf | 331KB | download |