We show that using time and location based features along with deep neural nets and transformers provide superior model performance compared to state of the art for Crime Prediction with 40% improvements of 5-10% higher accuracies.
Our code is in Crime ML.ipynb where you can see all the data preprocessing and machine learning parameters we used to replicate our results.
We obtained our data from OpenDataPhilly and the Federal Reserve Bank of St. Louis
The full dataset needed to run this project can be found here