Machine Learning Review

The lecture on machine learning was a new concept for me and rather difficult to follow. I have searched for machine learning and not found anything to help clarify but then searched for the individual components, ie K Nearest Neighbor (KNN) and had some better results.

 

My understanding of KNN is it's the easiest of the algorithms to classify data. The concept is to compare new data with test data which has already been analyzed by its attributes and given a class description. The idea of evaluating the distance is still a little vague but I can only imagine this is based on graphing somehow. The steps to determine KNN are the following:

 

1. Determine the number of nearest neighbors before making the analysis. This is 100% up to you to choose.
2. Calculate the distance of the new data from the trained data (the data already categorized)
3. Sort all of the distances of the K (the number of samples you chose) to the new data
4. Evaluate all of the class descriptions of the K values
5. Majority wins and this class becomes the class description of your new data set

 The KNN type of classification is considered “supervised” learning because you have already evaluated attributes and pre-classified the training data set. A good reference showing an example is located at: http://www.scholarpedia.org/article/K-nearest_neighbor

 

Support Vector involves drawing a line between groups of clusters and evaluating the margin distance of the data from the line. It was not clear how to determine where to draw the line or if a straight line is always draw, etc. This type of learning though seems to be common for speech recognition and hand writing analysis.

 

Clustering is another learning task where groups of data are combined together with similar attributes and then new data is compared to those groups. This is considered unsupervised learning because the groups of data have not be previously given a class label.

 

These concepts are really new and if these are to be understood in much more depth, it would be helpful to go over some more concrete examples in class.

 

 

Posted by :   Debbie Carter