University of South Florida
Optimizing Hidden Neurons in Multilayer Perceptrons with a Hybrid Feature Selection Approach for Classification Tasks
Abstract
dc:descriptionIdentifying the ideal number of hidden neurons in multilayer perceptron (MLP) networks remains an ongoing challenge in the field of machine learning. Our research presents an innovative strategy to address this issue by combining data preprocessing, feature selection by a hybrid method, and a methodical approach to MLP architecture design. The extracted features through feature selection help to construct MLP architecture and implement k-fold cross-validation along with early stopping mechanism for finding optimal number of hidden neurons. Our method's key innovation is its capacity to dynamically adjust the hidden neuron count while optimizing model performance. Through iterative refinement of the hidden layer structure and continuous evaluation of performance metrics, we identify the most effective network configuration for classification tasks. To validate our approach, we conducted tests on seven distinct classification datasets from the UCI machine learning repository and Kaggle. Notably, we achieved 95.07% accuracy on the breast cancer dataset using only two hidden neurons.
Degree
thesis:*- Name thesis:degree_name
- MS in Computer Science (M.S.C.S.)
- Level thesis:degree_level
- Thesis
- Discipline thesis:degree_discipline
- Computer Science and Engineering
- Grantor
- University of South Florida
- Year dc:date
- 2024
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Boyidapu, Susmitha
- Contributors dc:contributor
-
- Srinivas Katkoori, Ph.D.
- Shaun Canavan, Ph.D.
- Sriram Chellappan, Ph.D.
Subjects
dc:subject × 6Identifiers
dc:identifier.*- Repository record dc:identifier
- https://digitalcommons.usf.edu/etd/11172