Wednesday, November 18, 2009

Fuzzy reasoning & Artificial Neural Network

Interesting assignment for my AI paper. The first one asks for me to calculate someone's learning difficulty level, given his IQ level and his recent test score (education domain). Another one asks to predict the level of H1N1 risk of a patient, given his severity of these symptoms: fever, breathing difficulty, fatique and coughing.

Fuzzy reasoning is useful when you're dealing with 'high IQ', 'average IQ', 'low IQ', etc. How would you translate 'high' or 'average' into crisp numbers like more than 160 it's 'high IQ'. If it's more that 130cm it's 'average IQ'. Of course the numbers can't be used for everyone (I may have a different opinion on how much is 'high IQ'). And you can't use simple algorithm if and else like we normally would do. Fuzzy sets on the other hand allows you to have 160 falls under 'high IQ' and 'average IQ' at say, 0.8 and 0.6 (on a scale 0 to 1) each. Once you have defined all that, you have to have your fuzzy rules to work on. Given someone's IQ level and his test score, you can calculate his learning difficulty level and then maybe you can have this result to propose a new set of tests that match his level.

Artificial neural network emulates how the brain works, well sort of. It has the capability to learn and predict stuff. Like the H1N1 case, the model will learn to reach a very good predict whether the patient has high risk of being infected with H1N1 or not. It is much dependant on number of cases, the more cases you feed the model, the better the result will be.

Personally I think it's all mathematics. Everywhere you see calculations. Once you understand how it works, it's pretty easy (the data I worked on is not that much hahaha). In real world, the data set is huge, so data mining is actually intriguing, I think. Worth knowing.

In the risk of my classmates meniru, I'm going to upload what I have done so far, which I doubt that they'll find it here until much later. If you manage to get it here anyway, hey I really don't mind you looking around. Just be kind enough to give feedbacks there's mistakes in the doc :P.



  1. just miss the day i am in university. fuzzy logic is pretty easy actually. so is neural nets, if the lecturer not boring.....

  2. but bewarned data can be skewed heavily. and what sucks about neural network. it is a black box. hard to predict stuff