STE Model for Air Quality Prediction

Implement a novel and effective spatial-temporal ensemble model for air quality prediction.

1. dat：Data of the model.

   • air_bj.mat: Air quality and meteorology data of one year, sampled once an hour, from 35 monitoring stations in Beijing. Location information(longitude and latitude) of stations.

2. baselines: Baselines of our model.

   • Linear regression
   • Gaussian process
   • Regression tree
   • SVR, Neural network
   • Deep neural network
   • FFA model: The model proposed in [2].

3. lib: Some relevant functions of our STE model.

   • granger_cause: Granger causality. To discover the spatial correlation.
   • libsvm-master: SVM & SVR. Need to install.
   • myKmeans: K-Means and fc-Means.
   • myNN: Neural network & Stack auto encoder.
   • myLSTM: LSTM.

4. src: Proposed STE model.

   • T.m: Only temporal part.
   • ST.m: Only spatial and temporal parts.
   • SE.m: Only spatial and ensemble parts.
   • STE.m: Spatial-temporal ensemble model.

Reliance:

1. libsvm

Reference:

1. Prediction as a candidate for learning deep hierarchical models of data (Palm, 2012)
2. Y. Zheng, X. Yi, M. Li, R. Li, Z. Shan, E. Chang, T. Li, Forecasting fine-grained air quality based on big data, in: Proceedings of the 21th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, ACM, 2015, pp. 2267–2276.