Methods and systems for feature selection

A method for feature selection, comprising:

• a) creating a plurality of initial feature subsets from data;
• b) designing a desired feature subset tree based on one or more user inputs;
• c) inducing said initial feature subsets on training data to obtain induced feature subsets;
• d) sorting said induced feature subsets obtained in c) in order of performance to obtain sorted feature subsets;
• e) pruning said sorted feature subsets obtained in d) or h) based on said desired feature subset tree to obtain pruned feature subsets;
• f) merging said pruned feature subsets obtained in e) to form merged feature subsets;
• g) inducing said merged feature subsets obtained in f) on training data to obtain induced feature subsets;
• h) sorting said induced feature subsets obtained in g) in order of performance to obtain sorted feature subsets;
• i) repeating at least one of e), f), g), and h) until a desired user specified criterion is satisfied.

A method for feature selection, comprising:

• a) creating a plurality of initial feature subsets from data;
• b) inducing said initial feature subsets on training data to obtain induced feature subsets;
• c) sorting said induced feature subsets obtained in b) in order of performance to obtain sorted feature subsets;
• d) classifying said data with a feature subset having the highest performance;
• e) measuring a confidence level of correct classification for said data;
• f) comparing said confidence value with a threshold value;
• g) if said confidence value does not satisfy said threshold value:
• g1) merging said sorted feature subsets obtained in c) or g3) to obtain merged feature subsets;
• g2) inducing said merged feature subsets obtained in g1) on training data to obtain induced feature subsets;
• g3) sorting said induced feature subsets obtained in g2) in order of performance to obtain sorted feature subsets;
• g4) classifying said data with a feature subset having the highest performance;
• and
• h) repeating at least one of e), f), and g), until a desired user specified criterion is satisfied or until said confidence value is above said threshold value in f).

A method for segmenting a dataset comprising a plurality of data, comprising:

• a) creating a plurality of initial feature subsets from each of said plurality of data;
• b) inducing said initial feature subsets on training data to obtain induced feature subsets;
• c) sorting said induced feature subsets obtained in b) in the order of performance to obtain sorted feature subsets;
• d) classifying said plurality of data with a feature subset having the highest performance associated with each data;
• e) measuring a confidence level of correct classification for each data;
• f) comparing said confidence level for each data with a threshold value;
• g) selecting one or more data from said plurality of data having said confidence value below that does not satisfy said threshold value;
• h) merging said sorted feature subsets obtained in c) or j) of said one or more data having a confidence value that does not satisfy said threshold value to obtain merged feature subsets;
• i) inducing said merged feature subsets obtained in h) on training data to obtain induced feature subsets;
• j) sorting said induced feature subsets obtained in i) in order of performance to obtain sorted feature subsets;
• k) classifying said one or more data with a feature subset having the highest performance;
• and
• l) repeating at least one of e), f), g), h), i), j), and k) until a desired user specified criterion is satisfied or until all of said plurality of data have confidence values above said threshold value in e).

A computer readable medium comprising instruction being executed by a computer, the instructions including a software application for feature selection, the instructions for implementing the steps of:

• a) creating a plurality of initial feature subsets from data;
• b) inducing said initial feature subsets on training data to obtain induced feature subsets;
• c) sorting said induced feature subsets obtained in b) in order of performance to obtain sorted feature subsets, wherein induced feature subsets having similar performance are sorted using a heuristic technique;
• d) merging said sorted feature subsets obtained in c) or f) to obtain merged feature subsets;
• e) inducing said merged feature subsets obtained in d) on training data to obtain induced feature subsets;
• f) sorting said induced feature subsets obtained in e) in order of performance, wherein induced feature subsets having similar performance are sorted using a heuristic technique;
• and
• g) repeating at least one of d), e), and f) until a desired user specified criterion is satisfied.

A computer readable medium comprising instruction being executed by a computer, the instructions including a software application for feature selection, the instructions for implementing the steps of:

• a) creating a plurality of initial feature subsets from data;
• b) designing a desired feature subset tree based on one or more user inputs;
• c) inducing said initial feature subsets on training data to obtain induced feature subsets;
• d) sorting said induced feature subsets obtained in c) in order of performance to obtain sorted feature subsets;
• e) pruning said sorted feature subsets obtained in d) or h) based on said desired feature subset tree to obtain pruned feature subsets;
• f) merging said pruned feature subsets obtained in e) to form merged feature subsets;
• g) inducing said merged feature subsets obtained in f) on training data to obtain induced feature subsets;
• h) sorting said induced feature subsets obtained in g) in order of performance to obtain sorted feature subsets;
• i) repeating at least one of e), f), g), and h) until a desired user specified criterion is satisfied.