APPLICATION OF EXTREME LEARNING MACHINE AUTOENCODER FOR FINANCIAL TIME SERIES DATA

Abstract

This paper explores the application of Extreme Learning Machine
Autoencoder (ELM-AE) for the prediction of financial time series data, with a focus
on cryptocurrency markets. Extreme Learning Machines (ELMs) offer fast training
speeds and strong generalization capabilities, making them well-suited for modeling
complex, non-linear financial data. By integrating an autoencoder architecture, the
model effectively extracts meaningful features from noisy and high-dimensional
financial datasets. We present a practical implementation of ELM-AE for predicting
Bitcoin price movements, leveraging historical price data, technical indicators, and
relevant market features. The proposed model is evaluated using performance metrics
such as Mean Squared Error (MSE) and directional accuracy, demonstrating its
effectiveness in capturing market trends. Additionally, we develop a software
prototype that visualizes the prediction process, providing an interactive
demonstration of Bitcoin price forecasting.