0
$\begingroup$

I'm a Phd student and I have the results of some approaches (algorithms) that I would like to analyze. Data (results) are stored in csv files as follows: - the lines describe each algorithm with its parameters and the result obtained. - Some columns (characteristics) are quantitative and others are qualitative (enumerated types) and the last column is the results which presents the efficiency of the algorithm and it is a numerical value that I can convert in classes (of intervals of values). I would like to highlight at first what is the tendency of the good and worse algorithmes. Then, what are the features that contributed to obtain these results I'm beginner in machine learning and I searched a lot on the net but i didn't find the appropriate method . Could you point me to a clustering/ learning method for this purpose. I would be very greatful. Best regards.

$\endgroup$
0
$\begingroup$

K-means algorithm cannot be directly used for data with both numerical and categorical values because of its objective function. K-means uses Euclidean distance, which cannot be defined for categorical data.

However, there is a simple method which combines Euclidean distance and Hamming distance for the finding the similarity between instances that include both numerical and categorical features. In this method Euclidean distance is used as a metric for computing the similarity between numerical data and hamming distance for computing the similarity between categorical data. Both of them with the appropirate weights form an objective function that can handle mixed data.

For more information check the relevant paper

https://grid.cs.gsu.edu/~wkim/index_files/papers/kprototype.pdf

$\endgroup$
  • $\begingroup$ thanks so much , i'll try it and come back if i need help. $\endgroup$ – sara zohra AHMED BACHA Feb 12 '19 at 10:11

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.