The objective of this study is to satisfactorily produce a botnet traffic detection model, using pre-processing, feature engineering and optimization techniques specifically for the CTU-13 dataset, which has real samples of malware related traffic in addition to normal and background traffic. The methodology, in short, was: removal of invalid data through simple imputation; encoding; grouping in 5 second windows, source address and label; evaluation of prediction results. For the final evaluation, the following were used: Autoencoder, Stacked Autoencoders, Variational Autoencoder, Random Forest and KNN. All models showed good metrics, and the best results were from Random Forest, with a 0.96 f1-score.