The scientific community has recognized the challenge of enzyme classification in bioinformatics. In this study, we present a classifier implementation based on the vector representation of sequences, anomalous autoencoders, and convolutional networks, which we use to characterize chitinases belonging to glycosyl hydrolases. We developed this implementation using the Python programming language and the TensorFlow framework. We compare the performance of the deep learning model with a model built using XGBoost. The classifier consists of three levels that determine whether a protein is an enzyme or a hydrolase and its enzymatic activity. This considers the low representativeness of these enzymes in the Cazy.org database. The results for the first two levels of the classifier were similar for the neural networks and the XGBoost model, with an accuracy of around 90%. However, at the third level, accuracy dropped to 81%. Additionally, the proteome of Bacillus spp. Was explored for potential enzymes in these classes, and the results were compared with those of ProteInfer. To interpret and evaluate the significance of the features, we applied the SHAP (SHapley Additive Explanations) framework to the predictions generated by the XGBoost model. Further improvements in accuracy, explanations, and biological interpretations could be obtained for hybrid machine learning and deep learning classification models for these biotechnology-useful chitinase enzymes.