Hi everybody,

I require some information regarding Data Mining algorithms. I am quite new to the area and don't know exactly which algorithm to use for the project I am working on.

I am trying to simulate a smart home environment which learns and adapts to the inhabitants' behavior.

So far I have a house, with appliances, lightning and so on and a user interacting with them. Using Data Mining and Artificial Intelligence Pattern Matching, I would like to create a User Profile; basically a set of rules which describe the behavioral patterns of the user.

Then, I use this User Profile in an Expert System to predict the user's behavior.

My question is, which data mining algorithm should I use to construct these behavioral patterns (rules)? I have seen the post regarding sequential patterns and sequential rules, but I don't know whether it could relate to my project.

Thank you!
Keith

Hi Keith,

Sequential rule/pattern mining algorithms are applied to a sequence database. A sequence database is a set of sequences of symbols. The goal of sequential pattern/rule mining is to find patterns that are common to several sequences. These algorithms could be applied to your data if you have several sequences.

For example, if you have several sequences of data from the same user and you want to find some rules/patterns that are common to several of these sequences, then you could apply such algorithm.

On the other hand, if you have only a single sequence of data for each user, then you may consider "episode rules". An episode rule is a rule that appear several time in the same single sequence, and appear within a maximum amount of time.

Besides this distinction, there are many variations of the basic algorithms that adds additional features. For example, for sequential pattern/rule mining, there are some variation of the basic algorithm that consider allow considering time constraints, symbols having different importance/weight, or even symbols having probabilities (uncertain data). There are also various measures for rules such as confidence, lift, etc.

Also, if instead of sequence of symbols, you have time series (list of numbers usually measured at regular interval), other algorithms can be applied.

So in my opinion, it depends a lot on the represenstation of your data.

By the way, I have previously worked on the problem of mining sequential patterns/rules from user data in e-learning systems to autommatically generate hint to users and also on mining rules from weblogs to to make webpage recommendation. If you are curious, here are some articles describing how we have solved the problem:

Fournier-Viger, P. Gueniche, T., Tseng, V.S. (2012). Using Partially-Ordered Sequential Rules to Generate More Accurate Sequence Prediction. Proc. 8th International Conference on Advanced Data Mining and Applications (ADMA 2012), Springer LNAI 7713, pp.431-442.

http://www.philippe-fournier-viger.com/sequential_rules_prediction_2012.pdf

Fournier-Viger, P., Nkambou, R & Mephu Nguifo, E. (2008), A Knowledge Discovery Framework for Learning Task Models from User Interactions in Intelligent Tutoring Systems. Proceedings of the 7th Mexican International Conference on Artificial Intelligence (MICAI 2008). LNAI 5317, Springer, pp. 765-778.

http://www.philippe-fournier-viger.com/fournier-viger_sequential_patterns_micai08.pdf

In the second article, we have made several extensions to the basic sequential pattern mining algorithm to deal with some specific needs in our data.

Hope this helps you a little bit,

Philippe



Edited 2 time(s). Last edit at 03/05/2013 11:31AM by webmasterphilfv.