Hi,
I have a dataset that contains lots of parameters and lots of records. The aim of my study is to provide implicit information from this dataset by using data mining and machine learning techniques.
In the first step, I applied PCA to reduce the number of parmeters and now I'd like to know what can I use as Artificial Intelligence techniques.
Thank you for your responses.
You may want to start with cluster analysis
http://en.wikipedia.org/wiki/Cluster_analysis
A linear and a rank correlation analysis of the features (= parameters) give also a lo of insights about redundanies etc.
Clustering will be the best I think. you may also can try k-means algorithm.
You can use any unsupervised learning algorithm. Clustering is one of the options. Kindly go through http://mlg.eng.cam.ac.uk/zoubin/papers/ul.pdf for further information.
Beside cluster analysis, you might want to look for relevant associations between variables and/or variable-value pairs in your dataset. Just check for 'pattern mining' or 'association rue mining' on Google scholar.
Entropy can be used to select discriminative features or eliminate uninformative features.
here are two paper regarding feature selection using entropy.
And here is regarding how to use entropy to measure interestingness
"Interestingness score
A feature is interesting if it tells you some useful piece of information. Because the definition of what is useful varies depending on the scenario, the data mining industry has developed various ways to measure interestingness. For example, novelty might be interesting in outlier detection, but the ability to discriminate between closely related items, or discriminating weight, might be more interesting for classification.
The measure of interestingness that is used in SQL Server Analysis Services is entropy-based, meaning that attributes with random distributions have higher entropy and lower information gain; therefore, such attributes are less interesting. The entropy for any particular attribute is compared to the entropy of all other attributes, as follows:
Interestingness(Attribute) = - (m - Entropy(Attribute)) * (m - Entropy(Attribute))
Central entropy, or m, means the entropy of the entire feature set. By subtracting the entropy of the target attribute from the central entropy, you can assess how much information the attribute provides."
http://msdn.microsoft.com/en-us/library/ms175382.aspx
http://msdn.microsoft.com/en-us/library/ms175382.aspx
- Badges
- Science topic
- Similar topics
- Computer Science
- Data Mining