I have a problem to obtain good performances with a dataset. I have to predict the flow of visitors in my city given the distance of origin of tourists and the number of inhabitants of their city of origin. I plotted a 3D graph and the distribution is very bad.

enter image description here

I first tried with Linear Regression, the simplest regression tool, I computed the MSE and the MAPE, with the MSE i noticed that the model didn't overfit, but in order to have another estimate, i computed also the MAPE with a mean absolute percentage error of 60%. So i decided to try with Random forest, i tried different number of estimators but the score does not increase. At the end, i decided to implement KNN, just to see how it performs, i computed the score for different values of k and i plotted the learning curves: enter image description here

So i'm not able to do better, i think that these data are very badly distributed. Someone could give me a hint?

  • $\begingroup$ Have you normalized the dimensions? It seems like a fairly easy problem by looking at your plot. The less the distance, the more the flow, perhaps not linear but quadratic or even exponential? $\endgroup$ – Tom Mar 3 at 13:32
  • $\begingroup$ What do you mean by normalizing the dimensions? No is not linear since linear regression performs very badly, and no, is not exponential since there are many outliers that does not fit the distribution. But yes, the less the distance the more the flow, even if there some outliers. Anyway with Linear regression KNN and Random Forest i obtain very bad results and i don't know what to do. $\endgroup$ – FraMan Mar 3 at 13:54

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.