The Amazon rainforest, the largest tropical biome in the world, plays an essential role in both society and global environmental balance. Through its vast biodiversity and carbon storage capacity, it also supports local cultures and provides resources for sustainable development. Deforestation prediction occupies a significant role mainly in monitoring, control, and conservation planning. The ability to predict where and when deforestation will occur allows authorities and organizations to take more effective preventive measures, allocate resources more strategically and develop policies that can mitigate negative impacts.Therefore, the study of methods to predict deforestation has been increasingly developed in recent years. This work aims to apply supervised machine learning methods and statistical methods, such as autoregression, LightGBM, and Long Short Term Memory (LSTM) neural network to predict multi-step deforestation in the Brazilian Legal Amazon, using past observations of deforestation and climatic variables from the region. The research carried out showed that the most efficient results were presented in models that used autoregression. Furthermore, the study showed good results for classifying and pre-pointing anomalies in the series, characterized by their high deforestation values, as well as the general patterns of the series. The states of Pará and Mato Grosso and the municipality of Apiacás presented better results related to the classification of peak points, showing an average F1-Score for the predicted steps of 83%, 90%, and 85%, respectively. By enhancing strategies for monitoring and controlling deforestation, this study has the potential to positively impact public policies, promoting a balance between economic development and environmental preservation and climate regulation.