Many leaf diseases that affect crop health cause severe mango farming concerns. This study employed deep learning techniques to analyze mango leaf disease categorization comprehensively. This study looks at the classification of five different mango leaf diseases using convolutional neural networks (CNNs): powdery mildew, dieback, gall midge, anthracnose, and healthy leaves. A dataset of 661 raw images was first gathered, and these raw images were augmented to create a final dataset of 2678 images. The study examines how preprocessing methods affect classification accuracy using five CNN models before and after data preprocessing. The outcomes show notable gains in accuracy following preprocessing, highlighting its critical function in improving model performance. Additionally, a new custom model is proposed and evaluated, demonstrating its effectiveness in reaching an impressive accuracy of 98.50%. This comparative study shows that the proposed model outperforms current CNN architectures, indicating that it has the potential to classify mango leaf diseases accurately. Our research aims to highlight the importance of preprocessing techniques and creating customized deep-learning models by presenting an effective method for automated mango leaf disease identification.