Cassava is an important food crop in a wide range of tropical and subtropical regions; however its productivity is badly affected by a number of leaf diseases including Cassava Bacterial Blight, Brown Spot and Green Mite infestation and Mosaic Disease. Early and accurate identification of these diseases is crucial for efficient crop management but the traditional diagnosis techniques are time-consuming and heavily rely on expert knowledge. In this paper, an automated framework a Deep Larning (DL) based framework is proposed for automated cassava leaf disease classification using several convolutional neural network architectures, including DenseNet121, InceptionV3, MobileNetV2 and a hybrid fusion model called NeuroFusionNet (EfficientNetB0 + MobileNetV2). Extensive data preprocessing and augmentation techniques is learned to better the robustness and generalization under the real world scenarios. Experimental results show that the overall accuracy of MobileNetV2 is the highest of 93.14 %, which is heavier than other heavier models and still has computational efficiency. Furthermore, Explainable AI (XAI) with Grad-CAM is adopted to visualize model attention to ensure that the predictions are due to biologically relevant regions of the disease. The proposed framework brings an accurate, interpretable, and lightweight solution for real-time application of agriculture.