1. Feature Selection for Pattern Recognition J.-S. Roger Jang ( 張智星 ) CSIE Dept., National Taiwan University ( 台灣大學 資訊工程系 ) http://mirlab.org/jangjang@mirlab.org Machine Learning Feature Selection

2. 2 Feature Selection: Goal & Benefits Feature selection Also known as input selection Goal To select a subset out of the original feature sets for better recognition rate Benefits Improve recognition rate Reduce computation load Explain relationships between features and classes

3. Machine Learning Feature Selection 3 Exhaustive Search Steps for direct exhaustive search 1.Use KNNC as the classifier, LOO for RR estimate 2.Generate all combinations of features and evaluate them one-by-one 3.Select the feature combination that has the best RR. Drawback d = 10  1023 models for evaluation  Time consuming! Advantage The optimal feature set can be identified.

4. Machine Learning Feature Selection 4 Exhaustive Search Direct exhaustive search x2x3x4x1 x1, x2 x1, x3 x1, x4 x1, x5 x1, x2, x3 x1, x2, x4 x1, x2, x5 x1, x2, x3, x4 x1, x2, x3, x5 1 input 2 inputs 3 inputs 4 inputs x2, x3...... x5 x1, x3, x4 x1, x2, x4, x5..................

5. Machine Learning Feature Selection 5 Exhaustive Search Characteristics of exhaustive search for feature selection The process is time consuming, but the identified feature set is optimum. It’s possible to use classifiers other than KNNC. It’s possible to use performance indices other than LOO.

6. Machine Learning Feature Selection 6 Heuristic Search Heuristic search for input selection One-pass ranking Sequential forward selection Generalized sequential forward selection Sequential backward selection Generalized sequential backward selection ‘Add m, remove n’ selection Generalized ‘add m, remove n’ selection

7. Machine Learning Feature Selection 7 Sequential Forward Selection Steps for sequential forward selection 1.Use KNNC as the classifier, LOO for RR estimate 2.Select the first feature that has the best RR. 3.Select the next feature (among all unselected features) that, together with the selected features, gives the best RR. 4.Repeat the previous step until all features are selected. Advantage If we have d features, we need to evaluate d(d+1)/2 models  A lot more efficient. Drawback The selected features are not always optimal.

8. Machine Learning Feature Selection 8 Sequential Forward Selection Sequential forward selection (SFS) x2x3x4x1x5 x2, x1 x2, x3 x2, x4 x2, x5 x2, x4, x1 x2, x4, x3 x2, x4, x5 x2, x4, x3, x1 x2, x4, x3, x5 1 input 2 inputs 3 inputs 4 inputs......

9. Machine Learning Feature Selection 9 Example: Iris Dataset Sequential forward selection Exhaustive search

10. Machine Learning Feature Selection 10 Example: Wine Dataset SFSSFS with input normalization 3 selected features, LOO RR=93.8% 6 selected features, LOO RR=97.8% If we use exhaustive search, we have 8 features with LOO RR=99.4%

11. Machine Learning Feature Selection 11 Use of Input Selection Common use of input selection Increase the model complexity sequentially by adding more inputs Select the model that has the best test RR Typical curve of error vs. model complexity Determine the model structure with the least test error Model complexity (# of selected inputs) Error rate Test error Training error Optimal structure