Developing a predictive analysis system for automatic detection and monitoring of deforestation is one of the leading causes of phenomena such as biodiversity loss and climate change. Africa has significantly lost its forest cover after a decade of high-level of deforestation. While there are many machine learning models that have been developed globally to detect deforestation using generalized datasets across the globe, these models may lack effectiveness due to the diverse nature of deforestation and their lack of specificity, which affects their accuracy in other countries. A solution based on the deep learning models trained on labeled datasets from Hansen et al.; “Global Forest Change 2000–2022” extracting data from Uganda only with the tile covering 0–10 \(^\circ \) North, 30–40 \(^\circ \) East, was chosen. Each of the cropped layers were divided into smaller tiles of 100 \(\,\times \,\) 100 pixels (30 \(\,\times \,\) 30 km) to create a dataset that was used in this paper to train different deep learning models including convolution neural network, U-net, and recurrent network among others. The evaluation metrics that were used include accuracy, precision, recall, and f1 score. Lime, shape, and class activation maps were used for the explainability and interpretability of the models; however, class activation map was selected and used because it has effective high-level activations for local visualization and explaining the final prediction effectively. CNN achieved an accuracy of 82.3% and specificity 99.9%, U-Net with accuracy of 69.9% and specificity of 99.5%, and recurrent network with accuracy of 100% and 0% specificity. U-Net proved to be the most efficient model. The main goal is to give an estimate of the extent of deforestation, the specific locations, and possibly also information useful in cases of investigation of illegal deforestation.