Clan-Based Cultural Algorithm for Feature Selection
2019
Feature selection is the process of selecting a subset of relevant features from datasets, which can then be used for building machine learning models. In this study we propose a novel multi-population Clan-based Cultural algorithm (CCA) for feature selection. CCA models the complex social interactions within a clan-based society. Each population or clan contains solutions that are closely related and are managed by the population's own local belief space. There is also a global belief space that is shared by clans creating a tiered belief space structure. The global belief space contains the best solution across all the clans and provides the algorithm with a local and global view of the search landscape. At given generations, individual solutions migrate to another clan, different from its source clan, and combines with the clans population to form a single mating pool. Crossover and mutation operators are then applied and the fittest solutions are selected. This ensures that the search is directed toward the global optimum. The proposed algorithm was tested on benchmark datasets and its performance was compared with other evolutionary based feature selection methods. Results show that CCA achieved the highest accuracy, AUC and F1 values of 99.3%, 99.2% and 99.1% respectively across most of the datasets. The proposed CCA model selected fewer features to achieve its highest performance values than alternative approaches, suggesting CCA is more effective as a feature selection model.
Keywords:
- Correction
- Source
- Cite
- Save
- Machine Reading By IdeaReader
21
References
3
Citations
NaN
KQI