Bat Algorithm (BA) has recently started to attract a lot of attention as a powerful search method in various machine learning tasks including feature selection. Feature selection is essentially a dimensionality reduction problem that aims at two key objectives; to shorten the computational time and to improve classification accuracy. This paper attempts to address the issue of high computational time in feature selection by proposing a natural extension to the BA called the Bat Algorithm with Memory (BAM). This algorithm is inspired from observations that natural bats must rely upon spatial memory in navigating familiar spaces with dimensions larger than a few meters due to their limited biosonar operating range. Using the same approach, bats in existing BA is extended with memory capability to enable them to navigate easily over familiar locations. To evaluate the proposed algorithm, a series of experiments were carried out on twelve datasets with different number of objects and attributes. Next, the experimental results were compared with the original version of BA. The results showed that BAM was able to deliver competitive classification accuracy with increased saving time ranging from 28% up to 95%. The results also demonstrated that the time saving was attributed to three characteristics, which are samples number, feature number, and dataset geography. Consequently, BAM is also more efficient with lower number of features and higher number of samples.
|Number of pages||11|
|Journal||Journal of Theoretical and Applied Information Technology|
|Publication status||Published - 10 Sep 2015|
All Science Journal Classification (ASJC) codes
- Theoretical Computer Science
- Computer Science(all)