In this article, we study the effectiveness of the artificial neural net- work in forecasting the next day’s stock value (continuous data) and the overfitting problem during training, due to the high noise and redundant information in the datasets; for that, we trained an algorithm on historical data (Open, Close, High and Low prices) and we fed it to various machine learning models: artificial neural networks (ANN), Recurrent Neural Networks (RNN). We evaluated the model’s performance using different metrics generally used for machine learning algorithms. However, we used another method to back-test these models and evaluate their effectiveness, this method is based on transforming the predicted values (prices) into two classes (BUY, SELL signal).