If we follow the prediction setting of previous papers, the point-of-interest (POI) features, spatial distance features, and taxi flow features should also be added to predict the target (crime / house price).
POI features
Use the POI features generated from foursquarePOI.py. Raw count of POI in each category is preferred over the normalized percentage.
Spatial distance features
Top 6 closest neighboring CAs should be considered. The reverse distance weight is applied on the target variable of nearby CAs.
An adjacency matrix of all CAs should be maintained. Each time a flip tract label actions may change this adjacency matrix.
Taxi flow features
Use the taxi flow features generated from taxiFlow.py
A pairwise taxi flow count matrix should be maintained.
If we follow the prediction setting of previous papers, the point-of-interest (POI) features, spatial distance features, and taxi flow features should also be added to predict the target (crime / house price).
POI features
Use the POI features generated from
foursquarePOI.py. Raw count of POI in each category is preferred over the normalized percentage.Spatial distance features
Top 6 closest neighboring CAs should be considered. The reverse distance weight is applied on the target variable of nearby CAs.
An adjacency matrix of all CAs should be maintained. Each time a flip tract label actions may change this adjacency matrix.
Taxi flow features
Use the taxi flow features generated from
taxiFlow.pyA pairwise taxi flow count matrix should be maintained.