Alzheimer’s disease is an initial phase of dementia which affects elderly individuals leading to cognitive deterioration and memory loss. The research focuses on leveraging deep learning algorithms for the early diagnosis of Alzheimer’s disease using an ADNI dataset. The dataset contains 14, 716 MRI images for training and 4,680 MRI images for testing, categorized into Mild Demented and Non-Demented classes. Initially, a deep learning model named Inception V3 was employed as a baseline, and various performance metrics such as Accuracy, Precision, Recall, F1-score, AUC, MCC, Cohen’s Kappa, Log Loss, Brier Score, Geometric mean, False Detection Rate, False Negative Rate are used for evaluation. To further improve diagnosis accuracy and robustness, this research proposes and evaluate an optimized model, AlzDeepNet- which is an enhanced version of Inception V3 by incorporating five dense layers and three dropout layers to prevent overfitting.The proposed model demonstrates superior performance over the baseline Inception V3 architecture, by achieving improvements across key metrics including 92.11% Precision (vs. 86.14%), MCC score of 0.75 (vs. 0.55), and a 0.2834 Log Loss (vs. 0.4664), making it a trustworthy approach for real-time Alzheimer’s disease diagnosis. The ROC curve analysis indicates a high AUC score, showcase the capability of the model to distinguish between mild demented and non demented cases with high confidence score. This model can be integrated into healthcare systems to assist radiologists in diagnosis of Alzheimer’s for improving timely diagnosis with high accuracy.