Artificial neural network training using a multi selection artificial algae algorithm

Murat Karakoyun

doi:10.1016/j.jestch.2024.101684

2024-05-01

Murat Karakoyun *

DOI：10.1016/j.jestch.2024.101684

原文链接

查看原文

摘要

En 中文

Artificial neural network (ANN), developed by modeling the nervous system of the human brain, is an important and effective data processing method used today. The most important and difficult process of the artificial neural network is the training process. The main purpose of the training process is to optimize the weights in the network. The fact that the number of weights increases depending on the number of connections in the neural network makes this problem difficult and complex. Many algorithms and approaches have been presented from past to present in order to overcome this problem. One of the approaches used recently for ANN training is meta- heuristic algorithms. In this study, the artificial algae algorithm (AAA), one of the meta-heuristic algorithms, was used for ANN training. By adding a new selection mechanism to the position update structure of the basic AAA, a new AAA variant named multi selection AAA (MsAAA) has been developed. The multi-selection mechanism provides AAA with different options during the position update process, enabling a more effective and highquality search. With this improvement, the success of the algorithm has been increased by reducing the risk of stuck into local best. The performance of the proposed algorithm was compared with the performance of the basic AAA and 7 different meta-heuristic algorithms. The experimental results obtained on 21 different data sets were presented comparatively with four different metrics such as sensitivity, specificity, precision and f1 -score. Additionally, the performance of the algorithms was compared one -to -one, statistically and with visual graphics. The experimental results have shown that the proposed MsAAA is quite successful and outperforms other algorithms.

Keyword:

Artificial algae algorithm

Artificial neural network

Meta-heuristic algorithms

Multi -layer perceptron

Optimization