I found that there is two main technique for feature selection,
1. Filter
2. Wrapper.
My question is that where particle swarm optimization , artificial bee colony , gray wolf optimization (Evolutionary algorithm) falls under Wrapper Feature selection method?
Second question is : Is it feasible to use PSO for distributed node on big data? ( i want to select subset of features from distributed node and then only selected feature will travel at server node)
I need to find out proper feature selection algorithm which should work better on different distributed node and gives me relevant features only .
Generally, feature selection is consisted of two main step: (1) the generation procedure that find subset feature, and (2) the evaluation procedure that measures the quality of the obtained subset feature (step 1). As you may guess, the first step of feature selection should be done based on a search strategy such as evolutionary algorithms which are efficient search strategy for for approximating best solutions. But, feature selection methods are classified into filter-based, wrapper-based, and embedded-based (the second step). For more details you can refer to the following paper:
Article Hybrid of binary gravitational search algorithm and mutual i...
Hamid Bostani , Thanks for reply , can you tell me evolutionary feature selection comes uder wrapper Feature selection method or not ?
Can i use particle swarm optimization for distributed node implementation in bigdata ?
As mentioned, evolutionary algorithms such as PSO is a heuristic search strategy, not a feature selection. In other words, you can use them in a filter-based feature selection methods. However, these kind of algorithms usually are used in wrapper-based feature selection methods. In addition, I cannot understand what you mean about the second part of your question. You should state more details.
To address Big Volume Data the Parallel, Forward–Backward with Pruning (PFBP) algorithm with hybrid partitioning PFBP may be suited and Bootstrap-based statistical tests may be performed. Please go through the materials of above.
Dear Suman,
Genetic algorithm with binary chromosomes (and many other heuristic search algorithms) is a very good tool for wrapper feature selection. There are plenty of papers on this problem. See for example:
Dudek G.: Tournament searching method to feature selection problem. In: Rutkowski L. et al. (eds.): Artificial Intelligence and Soft Computing, ICAISC 2010, LNCS 6114, pp. 437-444.
https://gdudek.el.pcz.pl/files/Tournament_Searching10.pdf
where I compare several algorithms.
Best regards,
Grzegorz
Just rephrasing Hamid Bostani 's answer with simpler words. Filter and Wrapper are just two different ways to evaluate feature subsets in a feature selection process. Therefore, Evolutionary Algorithms can be both Filter/Wrapper, it just depends on what objective (fitness) function you are using.
If your objective function make use of a Machine Learning algorithm, and evaluate each solution (i.e. feature subset) by the accuracy/error, etc., then it is considered as a Wrapper. To the best of my knowledge, most research papers are using this way.
If your objective function is based on Correlation, Consistency, Rough set theory, etc., then it is considered as a Filter. There are fewer papers in this domain (EA for Filter Feature Selection), perhaps u could look at the followings which use Harmony Search and Intelligence Water Drop algorithms:
https://ieeexplore.ieee.org/abstract/document/6204102
Article An Ensemble of Intelligent Water Drop Algorithm for Feature ...
- Badges
- Science topic
- Similar topics
- Mathematics