1. Residuals: difference between the prediction and the number of crimes in 2016. It gives a good overview of the predictive model accuracy and the areas where it's (relatively) less performant. The predictive model supporting these data got a 0.92 R2 score which is very good.
2. Clustering layer, this is a description of the crime type patterns per LSOA over the 4 last years (2013-2016). Outliers are LSOA with a large value of crime over the average regarding some type of crime.