Signal Prediction in Cryptocurrency Tradeoperations : A Machine Learning-Based Approach

Deciding the right time to purchase a cryptocurrency is a crucial factor in enhancing a return on a given investment. In this work, we propose the use of gradient boosting algorithms (XGBoost and LightGBM) to perform the prediction of a binary market entry signal. For that, we use as features, in addition to prices and trading volumes, some technical market indicators. We use PCA (Principal Component Analysis) to reduce the dimensionality of the training/test datasets and Bayesian optimization to tune hyperparameters of the classification models. We have verified that the resulting strategy presents better results than a simple buy-and-hold of cryptocurrencies in the portfolio