Recently, me and my students have been working on developing a system for learning user behavior on redeem and claim actions for optimizing marketing campaign of a mobile phone maker data. The data is quite big (500 thousands transactions and about 25 thousands users). In adopting machine learning techniques to process the data, we need to make many data vectorization to convert categorical attributes to numerical attributes.
Is there any good guide on making this conversion make sense and safely?
The first thing you should do is to figure out which categorical attributes are ordinal (e.g. "subjective preference") and which are nominal (e.g. "mobile phone brand"). Then...
Example for nominal variables:
Let's assume "Phone brand" has 5 possible values: "Nokia", "Samsung", "Apple", "LG" and "LTE". You should create a dummy binary variable for each possible value of the category (so, in this case, you'd have a "Nokia" variable, a "Samsung" variable and so on..), that can only take two possible values: 0 (if it doesn't belong to the category) or 1 (if it does).
Example for ordinal variables:
Let's assume a "Preference" variable has 4 possible values: "Hate", "Don't like", "Like", "Love". You should then attribute a numeric value to each possible value so that their order is preserved. So, one possibility would be... 1 is Hate, 2 is Don't like, 3 is Like and 4 is Love. Optionally, you can scale the variable so that the values are within a specific range (e.g. between 0 and 1, or between -1 and 1).
Good luck.
Hello
what's kind of item response in your categorical data ? (in some case you can use Lieckert scale )
The first thing you should do is to figure out which categorical attributes are ordinal (e.g. "subjective preference") and which are nominal (e.g. "mobile phone brand"). Then...
Example for nominal variables:
Let's assume "Phone brand" has 5 possible values: "Nokia", "Samsung", "Apple", "LG" and "LTE". You should create a dummy binary variable for each possible value of the category (so, in this case, you'd have a "Nokia" variable, a "Samsung" variable and so on..), that can only take two possible values: 0 (if it doesn't belong to the category) or 1 (if it does).
Example for ordinal variables:
Let's assume a "Preference" variable has 4 possible values: "Hate", "Don't like", "Like", "Love". You should then attribute a numeric value to each possible value so that their order is preserved. So, one possibility would be... 1 is Hate, 2 is Don't like, 3 is Like and 4 is Love. Optionally, you can scale the variable so that the values are within a specific range (e.g. between 0 and 1, or between -1 and 1).
Good luck.
Dear Bereder and Silva,
Thank you very much for your answer. I will try them.
- Badges
- Science topic