Selecting the correct deep learning architecture is a significant issue when training a new deep learning neural networks model. Even when all of other DL hyper-parameters are accurate, the selected architecture will define the final classification quality of the generated model. In our previous paper we described a unique classification methodology called ACKEM for efficient and automatic classification of data, based on an ensemble of multiple DL models and KNN input-based architecture selection. The ACKEM methodology does not restrict the classification to one specific model with one specific architecture, as a specific architecture might not fit some of the input data. The ACKEM methodology had a major constraint – it used a brute-force approach for selecting the most suitable K for its inner usage of the KNN algorithm. In this paper, we propose a genetic algorithm (GA) based approach, for selecting the most suitable K. This method was tested over multiple datasets including the Covid-19 Radiography Chest X-Ray Images Dataset, the Malaria Cells Dataset, the Road Potholes Dataset, and the Voice Commands Dataset. All the tested datasets served us in our previous work on ACKEM, as well. This paper proves that replacing the inefficient method of brute force with a GA approach can improve the ACKEM method’s complexity without harming its promising results.
|Title of host publication||Intelligent Computing - Proceedings of the 2021 Computing Conference|
|Publisher||Springer Science and Business Media Deutschland GmbH|
|Number of pages||12|
|State||Published - 2021|
|Event||Computing Conference, 2021 - Virtual, Online|
Duration: 15 Jul 2021 → 16 Jul 2021
|Name||Lecture Notes in Networks and Systems|
|Conference||Computing Conference, 2021|
|Period||15/07/21 → 16/07/21|
Bibliographical notePublisher Copyright:
© 2021, The Author(s), under exclusive license to Springer Nature Switzerland AG.
- Data mining
- Deep learning
- Ensemble classifier
- Genetic algorithm