We are building an ML application for forecast and anomaly detection in air particles pollution data in Moscow. Existing applications mostly display data from monitoring stations and city-level forecast, but our goal is to provide street-level forecast and anomaly detection. We hope this will provide a more detailed understanding of air quality and helps in identifying sources of air pollution.
We selected the bayesian model with the Gaussian process prior. The advantage of this type of model is the prediction of the target variable distribution, rather than a point estimate. It is important for anomaly detection and can be used for selecting locations for new monitoring stations.
In constructing the model, we rely on these articles:
Now the dataset is assembled and the main work is directly related to the construction of the model. After that, we plan to wrap the model in API and visualize the results on a web page.
