The model is predicting the number of 911 calls is dependent on latitudes and longitude...naturally, it would be expected that the number of 911 calls be strongly correlated with the amount of potential victims. ("The model score (R2) is 0.81, so its accuracy is significant" is an incorrect interpretation of the R2; it indicated the amount of variation explains by the model, which is much different from an accuracy metric, and the high value could be explained by the correlation)
I built a similar model which predicts the types of crimes in San Francisco using LightGBM (better than xgboost which is better than scikit-learn's GBMs/GBTs), filling lat/long, month, day-of-week, hour, and year (http://minimaxir.com/2017/02/predicting-arrests/). The classification aspect is much tricker than a simple regression. But even then, latitude and longitude constituted 70% of the Gain in the GBM model.
(as an aside, day-of-week/hour should likely be encoded as categorical variables using one-hot-encoding, although when I tested that in my post, the results were unchanged, oddly)
Minimaxir thanks for comment.
Regarding "accuracy" I agree that this word usage is unfortunate in case of R2. A "goodness" would be better description of R2.
Regarding the "one-hot" encoiding of inputs, I would say - it depends. I know it is a common practice to do "one hot" encoding in Machine Learning. There are cases when it really helps the model to converge (for example, sometimes Neural Networks would require such curation of input).
However, model that I have used in my post is Decision Tree based. And when you think for a moment about it, Decision Tree based model should handle numerical input just fine. Usually, Decision Tree based models do not even require standardization/normlization of input and work correctly without it.
It's possible to predict the location and quantity of 911 calls based on call data (accident time, description of emergency calls and geolocation data) using Regression analysis
3 comments
[ 256 ms ] story [ 68.0 ms ] threadI built a similar model which predicts the types of crimes in San Francisco using LightGBM (better than xgboost which is better than scikit-learn's GBMs/GBTs), filling lat/long, month, day-of-week, hour, and year (http://minimaxir.com/2017/02/predicting-arrests/). The classification aspect is much tricker than a simple regression. But even then, latitude and longitude constituted 70% of the Gain in the GBM model.
(as an aside, day-of-week/hour should likely be encoded as categorical variables using one-hot-encoding, although when I tested that in my post, the results were unchanged, oddly)
It's possible to predict the location and quantity of 911 calls based on call data (accident time, description of emergency calls and geolocation data) using Regression analysis