Transfer Learning Approaches for Improved Thyroid Detection
D. M. Kalai Selvi, Kireet Joshi, V. Vani, Daxa Vekariya, L. Natrayan, Harshal Patil
Abstract
Thyroid disease is a broad term for any medical disorder that impairs the thyroid gland's ability to generate enough hormones. It is not limited to any particular age group; it can strike anyone at any time. Detecting thyroid diseases usually includes blood tests, which can be difficult to interpret due to the large amount of data required for accurate predictions. To find the best algorithm for predicting thyroid risk, this study examines different Transfer Learning (TL) methods. The dataset utilized for this work was sourced from the machine learning repository at the University of California, Irvine. The distribution of thyroid class labels in this dataset is highly imbalanced. To address this, sampling techniques are used to balance the dataset. Transfer learning models like AlexNet, Inception, and MobileNet are employed. All models achieve an accuracy rate of 97%. Upon comparison, AlexNet demonstrates the highest accuracy among the three models. The research shows that proper preprocessing and data balancing can improve thyroid detection efficiency.