Related Experiment Videos
An incremental approach to genetic-algorithms-based classification
1Department of Electrical and Computer Engineering, National University of Singapore, Singapore 119260. eleguans@nus.edu.sg
Summary
This study introduces genetic algorithms (GAs) for incremental learning, enabling systems to adapt to new data over time. The proposed methods effectively integrate new attributes and improve classification rates compared to retraining.
Area of Science:
- Machine Learning
- Artificial Intelligence
- Evolutionary Computation
Background:
- Incremental learning addresses dynamic environments and evolving datasets.
- Existing research predominantly uses statistical or neural network methods.
- Evolutionary algorithms are under-explored for incremental learning tasks.
Purpose of the Study:
- To propose novel genetic algorithm (GA) based approaches for incremental learning.
- To enable classifier agents in multiagent systems to adapt to new attributes.
- To enhance classification performance in dynamic learning scenarios.
Main Methods:
- Four new GA-based incremental learning strategies were developed.
- An 'integration' operation combines old solutions with new attributes.
- Biased mutation and crossover operations refine evolved solutions.
Main Results:
- The proposed methods successfully integrate new input attributes.
- Simulation results demonstrate improved classification rates over retraining GAs.
- The approaches effectively handle evolving input spaces.
Conclusions:
- Genetic algorithms offer a viable approach for incremental learning.
- The proposed methods enhance adaptability and classification accuracy.
- Potential applications include continuous training and feature selection.