View Full Version : wich method to use?

04-16-2008, 09:49 AM

I'm having a problem to deside wich data mining method can bring a solution to my problem. First a little description about my data-base. This is a medical data-base written in my-sql. For my problem I'm using 3 columns: the first containts the number of the patient (each patient has a unique number), the second column containts the number of a medical drug that was given to the patient, the third column containts the start date (DATETIME) when the drugs was given to the patient.
I have to find a way to see if there are some drugs that most of the time gives the patients another sickness wich results in other medication on al later date.
I'm usin weka as data-mining program.
Wich dm method can i use? Association rules? than i have a problem with "can't handle numerical data".


04-16-2008, 04:36 PM

It doesn't look like you have enough attributes in your data to do anything useful. You will need to have some attributes that describe the illness/side-effect suffered by individual patients. Furthermore, Weka is not well suited to temporal data. The data will need to be reformulated to propositional form that encodes the temporal aspect (if necessary). It sounds like it might be possible to use association rule mining - perhaps class association rules (those which have the target attribute in the consequent of the rules) - on this problem. Each patient could become multiple instances in the data, where each subsequent instance represents all the drugs that had been administered up to that point in time along with the current sickness/side-effect present (if any).


04-25-2008, 02:01 PM
...thrown on fire, but:

1. Data Mining is not meant to be done by non-Staticians - IMHO - so, get the help of one. There is a lengthy discussion on this and I know most of the arguments, agree with lots of them, not with the other lot, so please, don't trash me with them, I am justing giving my opinion trying to help our friend here.

2. If you work in a hospital or in a medicine faculty, it must be easy to find a Statician to get help from.

3. Because you are no Statician, you might get pointed to a no-right direction, reach some result, see things working fairly well and feel like results are true or meaningful. So, be weary when trying something and be very careful on publishing any research results obtained without the help of a Statician.

IMHO, Data Mining is a true scientific issue and so demands for true scientists/specialists to be done and business man to be carried on. It is not like gathering some data to report on them, or fixing a Java Class to work the way you know you want.

AGAIN: I am not preaching my gospel or anybody's, or pretending to know the truth, etc. It is just a word of advice from somebody who have faced similar questions.

Best regards,