A comprehensive study on evolutionary algorithm‐based multilayer perceptron for real‐world data classification under uncertainty. Issue 1 (10th August 2018)

• Record Type: Journal Article
• Authors: Dash, Tirtharaj; Behera, Himansu Sekhar
• Is Part Of: Expert systems. Volume 36:Issue 1(2019)
• Page Start: n/a
• Page End: n/a
• Publication Date: 2018-08-10
• Subjects: artificial neural network -- data classification -- evolutionary algorithms -- optimisation -- uncertainty; Expert systems (Computer science); 006.33
• DOI: 10.1111/exsy.12327
• Languages: English
• ISSNs: 0266-4720

Abstract

Abstract: In the area of neurocognition, classification of data is one of the most important phases. Conventional biologically inspired neural network models such as multilayer perceptrons (MLPs) are capable of learning and generalizing from exemplary patterns and are considered to be a popular choice for many different classification tasks. However, in the area of cognitive research, there lies a certain degree of uncertainty in acquired data. This uncertainty may be regarded as fuzziness. In this work, an attempt has been made to classify data which are associated with certain uncertainty. The resulting model is named as "FMLP." Further, MLP sometimes suffers from local minima problem during the training phase. To overcome the problem of getting trapped in the local minima in error back propagation, three different population‐based evolutionary metaheuristics (genetic algorithm, particle swarm optimisation, and gravitational search) have been implemented for training the FMLP. The resulting models are evaluated for seven real‐world benchmark datasets, and it has been found that the implemented models could demonstrate exemplary performance for real‐world data classification problems under uncertainty.