Academia.eduAcademia.edu

Figure 3 – uploaded by Kostantinos Demertzis

See full PDFdownloadDownload figure
Table 3. Accuracy (ACC) and Performance Matrices (PM) of ELM after feature selection by PSO search method The final conclusion is that the proposed method has proven to be reliable and effi- cient and has outperformed at least for these datasets the other approaches.

Table 3 Accuracy (ACC) and Performance Matrices (PM) of ELM after feature selection by PSO search method The final conclusion is that the proposed method has proven to be reliable and effi- cient and has outperformed at least for these datasets the other approaches.

Related Figures (5)

Fig. 1. The structure of the proposed algorithm
Fig. 1. The structure of the proposed algorithm
Graph 1. Original Lagocephalus Family Dataset
Graph 1. Original Lagocephalus Family Dataset
Graph 2. Lagocephalus Family Dataset after application of SMOTE
Graph 2. Lagocephalus Family Dataset after application of SMOTE
Table 1. Accuracy (ACC) and Performance Matrices (PM) comparison between ELM, MLP, SVM and RF algorithms in usual_fish_dataset
Table 1. Accuracy (ACC) and Performance Matrices (PM) comparison between ELM, MLP, SVM and RF algorithms in usual_fish_dataset
Table 2. Accuracy (ACC) and Performance Matrices (PM) of ELM after resampling by Bootstrap with replacement method Due to the fact that the number of the used features was too high for all three data- sets (lago_family_dataset 582 independent parameters, usual_fish_dataset 519 and lago_fasta_dataset 558) several feature selection attempts were done for the reduction of the training time and for the enhancement of the generalization in order to avoid overfitting. The Particle Swarm Optimization was used to search for the optimal fea- ture subset. The assessment of each subset was done by considering the value of each subset which is based on the contribution and the degree of redundancy of each cha-
Table 2. Accuracy (ACC) and Performance Matrices (PM) of ELM after resampling by Bootstrap with replacement method Due to the fact that the number of the used features was too high for all three data- sets (lago_family_dataset 582 independent parameters, usual_fish_dataset 519 and lago_fasta_dataset 558) several feature selection attempts were done for the reduction of the training time and for the enhancement of the generalization in order to avoid overfitting. The Particle Swarm Optimization was used to search for the optimal fea- ture subset. The assessment of each subset was done by considering the value of each subset which is based on the contribution and the degree of redundancy of each cha-
580 California St., Suite 400
San Francisco, CA, 94104
© 2025 Academia. All rights reserved