Drug discovery is a complex and multi-phase process aimed at identifying new therapeutic compounds that can effectively treat diseases. To enable researchers to assess the effectiveness of various compounds in inhibiting biological targets, we propose an intelligent solution based on a Deep AutoEncoder for the prediction of pIC50 values based on INFLUENZA virus. Our method was constructed using the ChEMBL dataset specifically on INFLUENZA-related compounds to train and test different predictive models. The experimental results show that combining Deep AutoEncoder with machine learning algorithms as the regression model, significantly improves predictive performance. In particular, the AutoEn coder combined with Random Forest model achieved the best results with a Mean Squared Error of 0.3284 and a coefficient of determination of 0.7391, out performing traditional models such as Ridge Regression and PLS Regression. These results demonstrate the effectiveness of applying deep learning for feature extraction in improving drug activity prediction in antiviral research.