A comparison of Kolmogorov-Arnold Networks with other time series forecasting models

Abstract

This thesis aims to evaluate the potential of KANs for time series forecasting. For this purpose, two KANs, the standard KAN and the Reversible Mixture of KAN (RMoK), are compared to forecasting methods with a different architectural structure. These are the deep learning models N-BEATS, N-HiTS, and LSTM, as well as the statistical models ARIMA and SARIMA.