Alzheimer’s disease (AD) prediction in neurology is a crucial task, and it requires enhanced predictive models for early identification and patient diagnosis. This study demonstrates the importance of outlier handling in biomarker data for improving the survival predictions for patients. This work used the box-plot method to handle and remove the outliers from biomarker data. After dealing with the outliers, Principle Component Analysis (PCA) was used to reduce the number of dimensions, and in later steps, wrapper-based feature selection was carried out. Furthermore, six different Deep Learning (DL) algorithms, which contain Multi-Layer Perceptron (MLP), Convolutional Neural Network (CNN), 3D CNN, 18 layers CNN, DenseNet149, and ResNet50, and five differentMachine Learning (ML) strategies, which contain Decision Tree (DT), Random Forest (RF), Gradient Naive Bayes (GNB), Logistic Regression (LR), and Support Vector Machine (SVM), were applied. We developed an optimized deep-learning survival analysis model based on the 18-layer Convolutional Neural Network (CNN) structure. We checked the proposed model against the dataset to identify potential factors influencing the progression of AD dementia. Regarding biomarkers’ data without outliers, the proposed 18-layered CNN method offered the best accuracy at 98% compared to the other models. The 18-layer CNN model with no outliers in the clinical biomarkers of Alzheimer’s data made good predictions. This study examined five factors that can cause memory loss, including age and three biomarkers (TNF, PGJ2, and NGF), and found that all five considerably raised the risk of AD dementia. Using the 18-layer CNN deep learning method, clinicians could most accurately predict the likelihood of a patient developing clinical AD dementia, including outliers, without relying on biomarkers.