Title: A method to evaluate task-specific importance of spatio-temporal units based on explainable artificial intelligence
Big geo-data are often aggregated according to spatio-temporal units for analyzing human activities and urban environments. Many applications categorize such data into groups and compare the characteristics across groups. The intergroup differences vary with spatio-temporal units, and the essential is to identify the spatiotemporal units with apparently different data characteristics. However, spatio-temporal dependence, data variety, and the complexity of tasks impede an effective unit assessment. Inspired by the applications to extract critical image components based on explainable artificial intelligence (XAI), we propose a spatio-temporal layerwise relevance propagation method to assess spatio-temporal units as a general solution. The method organizes input data into an extensible three-dimensional tensor form. We provide two means of labeling the spatio-temporal tensor data for typical geographical applications, using temporally or spatially relevant information. Neural network training proceeds to extract the global and local characteristics of data for corresponding analytical tasks. Then the method propagates classification results backward into units as obtained task-specific importance. A case study with taxi trajectory data in Beijing validates the method. The results prove that the proposed method can evaluate the task-specific importance of spatio-temporal units with dependence. This study also attempts to discover task-related knowledge using XAI.
The program of the ST-LRP is developed based on the previous work of LRP in GitHub (TensorFlow LRP Wrapper) (Lapuschkin et al. 2016).
Lapuschkin, S., et al., 2016. The LRP toolbox for artificial neural networks. The Journal of Machine Learning Research, 17 (1), 3938-3942.
- The file "ST-LRP.zip" is the zip package of the main programs in this study. All the programs are the Jupyter Notebook files in python (.ipynb):
The file "Holiday_revise_2019.ipynb" is the program of neural network model training and the spatio-temporal unit assessment.
The file "Validation1_revise_Replaced zones.ipynb" is the program of the first validation experiment (replace the zone value of input).
The file "Validation2_Regression_POI.ipynb" is the program of the second validation experiment (interpretation of unit importance measure difference based on POI data).
The file "Validation3_Random forests.ipynb" is the program of the third validation experiment (use random forests to assess units).
The file "Validation4_Cluster.ipynb" is the program of the fourth validation experiment (data compression application).
- The file "Trained Parameters.zip" is the zip package of trained neural network parameters.
The data that support the findings of the study are available in figshare. The data include the training set, validation set, and test set of the model training (i.e., taxi origin point distributions collected in the years of 2016 and 2017) and the inputs for the subsequent validation experiments.
Please consider citing our paper if this helps in your work:
Ximeng Cheng, Jianying Wang, Haifeng Li, Yi Zhang, Lun Wu, & Yu Liu (2020). A method to evaluate task-specific importance of spatio-temporal units based on explainable artificial intelligence. International Journal of Geographical Information Science. 1-24. DOI