Využití umělé neuronové sítě pro predikci finančního trhu

Abstract

This diploma thesis focuses on forecasting the movement of currency pairs using neural networks. The predictive models are based on the LSTM and N-HiTS architectures. The models were implemented in Python using the NeuralForecast library. Data for the EUR/USD and NZD/JPY currency pairs were downloaded from a publicly available source. The data were subsequently cleaned, resampled, enriched with technical indicators, and split into training and test sets. Forecasting was performed using a sliding window method and evaluated using forecasting accuracy metrics. The results were further verified by backtesting a trading strategy. The findings show that neural networks have potential for financial time series forecasting. However, their practical applicability depends on the quality of input data, model design, and the availability of computational resources.