The killer cancers of the world are lung cancer and colon cancer while considering the few deviations in tissues by histopathological examinations it is necessary to find the way out in earlier stages if it can be found which is not not easy. This research systematically assesses the performance of four state-of-the-art models working with Deep Convolutional Neural Network (D-CNN), DenseNet121, ResNet152V2, InceptionV3, and SE-ResNet152-based systems for the automated analysis of histopathology of lung and colon cancer. To foster transparency and drive clinician trust, we combined it with Grad-CAM, a method for explainable AI (XAI), to identify important areas that impact the model's predictions, and to maintain the interpretability throughout the training and evaluation processes. Using the LC25000 dataset under constant experimental settings, we have shown that the performance of SE-ResNet152 is better than the other models, and the results achieved an area under the curve (AUC) of 0.996 and an outstanding accuracy of 99.37%. This enhanced performance is an indication of its ability to distinguish malignant from benign tissues in spite of complex histological differences. Our proposed technique is not only able to enhance automated cancer diagnosis in a novel way, but also gives an interpretable and reliable framework that can be used in real-world clinical contexts.