Hybrid Particle Swarm Optimization with Spiral-ShapedMechanism for Feature Selection
Ke Chen, Feng-Yu Zhou*, and Xian-Feng Yuan
Abstract: The “curse of dimensionality”is one of the largest problems that influences the quality of the optimization processin most data mining, pattern recognition, and machine learning tasks. Usinghigh-dimensional datasets to train a classification model may reduce thegeneralization performance of the learned model. In addition, highdimensionality of the dataset results in high computational and memory costs. Featureselection is an important data preprocessing approach in many practicalapplication domains that are relevant to expert and intelligent systems.Feature selection aims at selecting a subset of informative and relevantfeatures from an original feature dataset. Therefore, using a feature selectionapproach to process the original data prior to the learning process isessential for enhancing the performance on the classification task.In this paper, hybrid particle swarmoptimization with a spiral-shaped mechanism(HPSO-SSM) is proposed for selecting the optimal feature subset forclassification via a wrapper-based approach. In HPSO-SSM, we make threeimprovements: First, a logistic map sequence is used to enhance the diversityin the search process. Second, two new parameters are introduced into theoriginal position update formula, which can effectively improve the positionquality of the next generation. Finally, a spiral-shaped mechanism is adoptedas a local search operator around the known optimal solution region. For acomplete evaluation, the proposed HPSO-SSM method is compared with sixstate-of-the-art meta-heuristic optimization algorithms, ten well-knownwrapper-based feature selection techniques, and six classic filter-basedfeature selection methods. Various assessment indicators are used toproperly evaluate and compare the performances of these approaches on twentyclassic benchmark classification datasets from the UCI machine learningrepository. According to the experimental results and statistical tests, thedeveloped methods effectively and efficiently improve the classificationaccuracy compared with other wrapper-based approaches and filter-basedapproaches. The results demonstrate the high performance of the HPSO-SSM methodin searching the feasible feature space and selecting the most informativeattributes for solving classification problems. Therefore, the HPSO-SSM methodhas broad application prospects as a new feature selection approach.
Keywords: Particle swarm optimization;feature selection; classification; optimization
Highlights:
HPSO-SSM is proposed based on originalparticle swarm optimization.
A new wrapper-based feature selectionapproach based on HPSO-SSM is proposed.
The logistic map sequence is used toenhance the diversity in the search process.
An innovative position update model ispresented to improve the position quality.
Our method outperforms seventeen extremelycompetitive methods in terms of accuracy.
Expert Systems With Applications (Top期刊, JCR一区, SCI二区,IF=3.768)