Use the crime prediction technique for another task -- predict average income

[thekingofkings]

To facilitate the argument that our regression model is able to solve other urban problems, we need another prediction task to evaluate.

Take the average income of CA prediction as example.

[thekingofkings]

The average income plot. The income standard deviation is plotted against the mean for each CA. The annotation is the CA ID.

As shown in the figure below, the income std is proportional to the mean. The downtown area usually have higher income level. ca-income.pdf

[thekingofkings]

The crime rate vs. average income plot.

As shown in the plot. Overall, the crime rate has a linear correlation with the average income. It is noisy though. ca-crime-income.pdf

[thekingofkings]

The taxi flow are from Year 2013. The average income is from census data in 2010.

Setting	MAE	MRE
NBR (demo + poi)	0.253	15304
NBR (demo + poi + taxi + geo) use volume in lag variable, e.g. KDD16	0.250	15127
NBR (demo + poi + taxi + geo) use MF embedding in lag variable	0.2752	16648
NBR (demo + poi + taxi + geo) use LINE embedding in lag variable	0.2567	15534
NBR (demo + poi + taxi + geo) use DGE embedding in lag variable	0.2436	14740