| Abdulvahit Karail, Yasin Ortakcı Impact of Optimizer: Comparative Analysis of Loss Functions in Different Models |
|---|
| Abstract. This study compares the performance of different artificial neural network models for data classification problems. The models used include Multilayer Perceptron (MLP), Kolmogorov–Arnold Network (KAN), and Liquid Time-Constant Networks (LTC). Model training was performed on the MITBIH ECG dataset, and modern optimization strategies with different loss functions were systematically compared. Evaluation metrics such as accuracy, macro-F1, precision, and recall were assessed. The findings show that KAN models provide high accuracy, while LTC models operate more slowly due to their time-step based calculations. |
| Keywords: Optimizers, loss functions, Kolmogorov–Arnold Network, Liquid Time-Constant Networks, Multilayer Perceptron |
 Download PDF |
| DOI: https://doi.org/10.54381/itta2026.1.07 |