TRANSFER LEARNING APPROACHES FOR IMPROVING MACHINE LEARNING MODEL ON SMALL DATASETS: CASE STUDY SKIN LESION DETECTION
No Thumbnail Available
Date
2024
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
One of the biggest challenges of working with machine learning is the need for large
amounts of data to train accurate models. Small datasets, in particular, can pose significant
challenges to machine learning models. Researchers have developed several approaches to
address the challenges of working with small datasets, including transfer learning, active
learning, and data augmentation.
In this study, we employed the transfer learning as a machine learning paradigm that
leverages knowledge gained from one task or domain (source domain) to improve
performance on a different but related task or domain (target domain).
This study focuses on optimizing model accuracy, analyzing feature importance, and
understanding the impact of different preprocessing techniques. The results demonstrate
significant improvements in diagnostic performance compared to traditional methods,
highlighting the potential of deep learning in clinical settings. Future work involves
enhancing model generalization and exploring the integration of multimodal data to further
refine diagnostic capabilities.
We started from VGG16 as a pre-trained model for training a baseline model in a CNN on
the domain of skin lesion detection.
In the experimental part, we evaluated different structures, including a custom CNN and
VGG16, in terms of detection accuracy. The results showed that transfer learning with the
pre-trained model VGG16 provided the best performance, achieving an accuracy of 85%,
highlighting the importance of transfer learning.