.png){kind=icon}
.png){kind=icon}
Return
Three-stage multi-objective feature selection for distributed systems
.png){kind=icon}
PRE
AI
DOI：10.1007/s00500-023-07865-y.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Deep learning and machine learning researchers must overcome the difficulty of big data analytics. One such technique for big data analytics is distributed systems, which allows data scientists to greatly boost their efficiency by (1) distributing concurrent experiments across multiple devices and (2) greatly lowering training time by sharing a single network's training across numerous devices. A learning approach requires a considerable amount of time and the effectiveness of the model degrades in large datasets because of redundant features. We employ the feature selection technique to pick a subset of pertinent and non-redundant characteristics to get over these issues. However, the majority of feature selection techniques are unstable, meaning that they choose various subsets of features for various training datasets, leading to varying degrees of classification accuracy. As a solution to this problem, a new three-stage multi-objective feature selection (TMFS) technique is proposed in this paper to improve the performance of distributed systems. By merging several subsets of features, the proposed TMFS technique chooses an optimal subset of features. This proposed TMFS algorithm concentrated on achieving multiple objectives, particularly three objectives. They are (1) Reduce the rate of classification error, (2) Decrease the features count, and (3) Reduce the dataset. The TMFS technique uses 5 feature selection strategies (Correlation coefficient, Fisher score, Information gain, Mean absolute deviation, and Min-max normalization) in 3 stages to accomplish these objectives. A Higgs Boson dataset and three machines were used to assess the TMFS algorithm at distributed systems. The test result shows that TMFS algorithm gives accuracy of 0.911, 0.916, and 0.842 for machine 1, machine 2 and machine 3, respectively.
Keywords:
TMFS
Big data
Feature selection
Journal
IF:
2.5
Papers:
1.0W
Citations:
2.1W