|
|
 |
| |
|
|
|
Hybrid Electric Eel Foraging Optimization and Quadratic Interpolation Optimization for Multilayer Perceptron Training |
|
|
|
PP: 1019-1027 |
|
|
doi:10.18576/amis/200412
|
|
|
|
Author(s) |
|
|
|
Imad El Karkri,
Abdelghni Lakehal,
|
|
|
|
Abstract |
|
|
| Multilayer perceptron (MLP) training is a difficult optimization task due to the complexity of searching for suitable weights and biases. Classical backpropagation may suffer from slow convergence, sensitivity to initialization, and stagnation in local optima. To address these issues, this paper proposes a new hybrid metaheuristic method, called EEFO-QIO, for MLP training. The proposed approach combines Electric Eel Foraging Optimization (EEFO) and Quadratic Interpolation Optimization (QIO), benefiting from the exploration capability of EEFO and the exploitation strength of QIO, while a restart mechanism is used to reinitialize candidate solutions when stagnation occurs, except for the global best solution. The method is applied to optimize the weights and biases of the MLP on five benchmark classification datasets: XOR, Balloon, Iris, Breast Cancer, and Heart. Its performance is evaluated using mean squared error, classification accuracy, and the Friedman test. The results show that the proposed EEFO-QIO method provides robust and competitive performance for MLP training. |
|
|
|
|
|
|
|
