Preference Learning Toolbox

Automatic Feature Selection

This toolbox supports a number of feature selection algorithms to automatically find the most relevant subset of input features for the model.\n + Different search methods can be selected to traverse the space of possible feature combinations. The performance of each subset of features considered is computed as the prediction accuracy of a model trained using that subset of features as input. All of the preference learning algorithms implemented in the tool can be used to train this model.

Search algorithm

The user can choose between two hill-climbing algorithms: sequential forward feature selection [1] and n-best individuals selection [1].

Evaluation method

The user can select the training algorithm and the method to evaluate its performance.

Training algorithms

Two training algorithms for multi-layer perceptrons[2] and one for support vector machines[3] are supported, namely backpropagation[4], neuroevolution[5] and ranking SVM[6].