An Ontological Approach for Recommending a Feature Selection Algorithm

Abstract

AbstractFeature selection plays an important role in machine learning or data mining problems. Removing irrelevant features increases model accuracy and reduces the computational cost. However, selecting important features is not a simple task as one feature selection algorithm does not perform well on all the datasets that are of interest. This paper tries to address the recommendation of a feature selection algorithm based on dataset characteristics and quality. The research uses three types of dataset characteristics along with data quality metrics. The main contribution of the work is the utilization of Semantic Web techniques to develop a novel system that can aid in robust feature selection algorithm recommendations. The system’s strength lies in assisting users of machine learning algorithms by providing more relevant feature selection algorithms for the dataset using an ontology called Feature Selection algorithm recommendation based on Data Characteristics and Quality (FSDCQ). Results are generated using six different feature selection algorithms and four types of classifiers on ten datasets from UCI repository. Recommendations take the form of “Feature selection algorithm X is recommended for dataset i, as it performed better on dataset j, similar to dataset i in terms of class overlap 0.3, label noise 0.2, completeness 0.9, conciseness 0.8 units". While the domain-specific ontology FSDCQ was created to aid in the task of algorithm recommendation for feature selection, it is easily applicable to other meta-learning scenarios.KeywordsFeature selection algorithmsMeta-featuresOntology