Розробка систем підтримки прийняття рішень на основі машинного навчання з мульти-таймфреймними даними для алгоритмічної торгівлі на ринку Форекс

The effectiveness of decision support systems (DSS) in the foreign exchange market (FOREX) critically depends on the selected timeframe and machine learning objective function (ML-objective). The need to determine the optimal combination of these factors determines the relevance of this study. The aim of the work is to develop and conduct a comparative analysis of the performance of three DSS architectures (Systems A, B and C) to establish the optimal mathematical formulation of the market forecasting problem. Three architectures were implemented: System A is based on regression forecasting; System B uses classification based on features of a single timeframe (15M); System C uses classification using multi-timeframe features. The models were trained on historical data, and their effectiveness was verified by backtesting on an independent (OOS) data set. The impact of multi-timeframe features and ML-objective on DSS performance for the XAU/USD asset was studied. According to the results of the comparative analysis, System C demonstrated exceptional results, surpassing other architectures: the total return was 3283.69%, and the maximum drawdown was only 2.07%. Prospects for further research include: developing an integrated risk management methodology based on ML models; implementing early position closing mechanisms (in particular, Trailing SL); and improving the DSS to compensate for model degradation.