Let me explain more about my question: I have collected 2000 data as the following:

age sex education residence music

young male Primary_school east mixture middle-agee female Primary_school north mixture young female Bachelor south_east mixture young male Master_degree east mixture old male physician east traditional middle-agee female Bachelor east mixture young female Bachelor center mixture teeneger female Primary_school west mixture young female Master_degree east mixture young female Master_degree south_east mixture young female Primary_school south mixture middle-agee female illiterate east mixture middle-agee male Bachelor south traditional young female Master_degree east mixture old male Primary_school south_east traditional middle-agee female Bachelor east traditional young female Master_degree south mixture ....

In our city, we have 22 districts with total population 12000,000 people. I have collected 2000 samples as the above from different locations by asking question from people each of the above question, like what is your age?, your education? your sex ? your district ( north, east, west,...) and what is your favorite music ( which kind of music do you like to listen)?

Now I want to use classification algorithm to predict if a new person selected from one area with specific age; sex; education; her residence (area), then we want to predict which kind of music she like to listen to?

I use R and examine 5 top algorithms such as svm, lda; knn; randon Forest and figured that the error is more than %70. [The OOB estimate of error rate is so high ~~%70]

Could you please explain why this bias and error happened?

Best regards Amir

  • 5
    $\begingroup$ There's no way for us to answer this, as you did not supply details on how you generated your data. For example, if you generated your predictors and classes independently, then you would expect any model to achieve at best an accuracy of 10%, so by that measure you are doing well. There is no absolute measure of a "good" error rate, it depends on the quality and quantity of the data you have. $\endgroup$ – Matthew Drury May 16 '17 at 15:23
  • $\begingroup$ Have you considered that there is some real noise in the data & that level of error is appropriate? $\endgroup$ – gung - Reinstate Monica May 16 '17 at 17:30

I think it's simple: because the information to make more accurate predictions is just not contained in the data. You cannot make more accurate inference, most probably.

I give you an example: I like to play guitar, study and have my own laptop. In which country do I live?

No chance to get that right...


Your Answer

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

Not the answer you're looking for? Browse other questions tagged or ask your own question.