Tools
Tools
Dhaygude, Pratik Anil (2024) Improving Image Classification Performance Through Advanced Deep Ensemble Techniques. Masters thesis, Dublin, National College of Ireland.
Preview |
PDF (Master of Science)
Download (3MB) | Preview |
Preview |
PDF (Configuration Manual)
Download (983kB) | Preview |
Abstract
Improvement of image classification forms the core of the study with refined deep ensemble strategies. Chapter 1 introduces by defining the problem, aim and objectives are defined as applied in enhancing the image classification accuracy, robustness, and explainability with the utilization of ensemble approaches.
Chapter two provides the literature review of image classification and deep ensemble, stacking, bagging, boosting and hybrid methods. Methodology is described in chapter three with an experimental design selected to facilitate the achievement of the research objectives. It outlines ways of generating and deploying the ensemble models as well as data pre-processing, model designs and training algorithms
Chapter four provides an overview of the process of applying and assessing the outcomes of employing different machine learning techniques for image classification using the “Flower” dataset. During EDA, class distributions and characteristics of images are found and the dataset is made more ready by performing data augmentation. Implementation of the model comes into CNN and some of the latest architecture such as ResNet 50, VGG 16, and Inception V3. Ensemble techniques used were stacking, Bagging, and Boosting. Other evaluation measures such as training and validation loss, accuracy, a confusion matrix, and classification report are given and the accuracy level varies from 68% to 94%. The presented results allow identifying the success rates of various models and techniques in image classification.
Chapter 5 compares the performance of different machine learning algorithms when applied to the ‘’Flower’’ dataset. CNN had the highest accuracy of 68% but the model seemed to overfit, and therefore regularization is needed. For feature extraction, the pre-trained models including ResNet50, VGG16, and InceptionV3 expressed good results. Thus, Stacking, Bagging, and Boosting were used and comparing the results we can state that Bagging has higher precision and F1-score. the percentage of classification accuracy remained rather moderate.
The last chapter demonstrated that ensemble models have helped enhance learners’ image classification accuracy. These objectives were met by creating and fine-tuning these techniques to show that they can indeed improve the model’s stability and performance. It has been reported that future advancements should also use more significant datasets, develop efficient computational strategies for effective execution of the models and enhance the readability of the same for better referencing or application.
| Item Type: | Thesis (Masters) |
|---|---|
| Supervisors: | Name Email Haycock, Barry UNSPECIFIED |
| Subjects: | Q Science > QA Mathematics > Electronic computers. Computer science T Technology > T Technology (General) > Information Technology > Electronic computers. Computer science Q Science > QA Mathematics > Algebra > Algorithms > Computer algorithms Q Science > Q Science (General) > Self-organizing systems. Conscious automata > Machine learning |
| Divisions: | School of Computing > Master of Science in Data Analytics |
| Depositing User: | Ciara O'Brien |
| Date Deposited: | 15 Aug 2025 17:28 |
| Last Modified: | 15 Aug 2025 17:28 |
| URI: | https://norma.ncirl.ie/id/eprint/8552 |
Actions (login required)
 |
View Item |