ADMEWORKS ModelBuilder - Feature Selection - Boost-LDA FS
This command displays a dialog for Feature Selection using Ada-Boost algorithm. It is a forward algorithm of feature selection. It starts with an empty parameter set. At each iteration it chooses one parameter, adds it to the parameter set and builds a weighted FLD model. After each iteration it makes a decision, whether to add a new feature in the next iteration. In this decision it takes into account the following conditions: parameter min. distance and accuracy (in the same window length number) in previous iterations. The algorithm ends when the number of iterations reaches the max. number of iterations, or when the FLD models can not improve (because the percentage of correctly classified samples is 100%).
The algorithm works best when the data set does not contain zero values, or correlated parameters. Thus before running the Boost-LDA feature selection, execution of the Correlation Test and/or Multicollinearity Test is required.