Realized Tasks:
Forecasting oil prices taking into account geopolitical, climatic and other difficultly predicted factors.
Implementation path:
To increase the accuracy of the forecast, it was decided to use a combination of several methods, mutually exclusive of each other’s shortcomings, when building the model. As a result, we get a hybrid model that, taking into account the requirements for the final product, most fully reflects the complex relationship of various factors. Econometric models were used to predict individual factors, as well as the C-SVC support vector method in combination with the AdaBoost M1 boosting algorithm. For the final forecasting of oil prices, a recurrent neural network LSTM was built on the basis of the results of econometric models.
Tools used:
R, C ++, Python (Scikit-learn, Keras, Matplotlib libraries)
Achievements:
Project completed