0
$\begingroup$

I am new at statistics and ML. Due to my lack of theoretical background I was wandering if does it make sense to combine NBC and CC. I am participating to the kaggle competition https://www.kaggle.com/c/sf-crime and I have to classify a crime on the basis of a data set composed of those categories predict the type of crime:

(Dates,DayOfWeek,PdDistrict,Address,X,Y) ----> Type of crime

I would like to dump the X Y features and implement with the remaing a NBC. Then to capture the correlation among the variables create a graph for each crime category: the nodes would be the rows of the NBC and the weight of the links which represent occurrences in which those factors appeared together.

For example (Wednesday, prostitution 12.00 ingleside) will increase by one all the links among those nodes.

Once this is done I would like to compute the CC for each node and try for each new crime how it fits in those graphs summing up all the CC of the respective node. I would like to combine this result with the classification of NBC.

Does it makes sense? Does I am reinventing the wheel? If this method makes sense, is there a way to do it?

$\endgroup$
0
$\begingroup$

You might be better off concentrating on use of additional classifiers instead of only focusing on only NBC. For example, also use kNN, multivariate linear regression, linear discriminant analysis, etc. You seem to be hooked on one classifier. Advanced classification techniques implement "ensemble classifier fusion," or the combining of votes from multiple classifiers, and on the advanced end of the spectrum address the "diversity" of multiple classifiers (Kuncheva et al). NBC alone is not be sufficient and will not be robust, that is, will not perform well in many environments on different datasets.

$\endgroup$
1
  • $\begingroup$ Yep I get it, but I have to do three university's exams from here to 1 February: I have very thigh deadlines! You know how are does things I want everything under control, I can study up to 16 house per day, that's not a problem, but I need focused material.... do you have any focused link or anything else that you would use in my position? $\endgroup$
    – Aalto
    Jan 16 '16 at 9:27

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.