<p>AI algorithms, in drug discovery, support target identification by recognizing biological patterns and molecular interactions linked to disease mechanisms. They further aid in lead compound optimization, virtual screening of large chemical libraries, and do drug design, thereby reducing time and cost constraints traditionally associated with laboratory-based approaches. This research proposes a novel framework that integrates the Framingham Heart Study (FHS)—a gold standard longitudinal dataset in cardiovascular research—with advanced machine learning (ML), deep learning (DL), and explainable artificial intelligence (XAI) techniques to predict the risk of death or survival probability based on cardiovascular risk factors and to enhance pharmaceutical marketing precision. The study leverages structured data from FHS, encompassing risk factors such as age, cholesterol levels, blood pressure, smoking status, and diabetes incidence, to model predictive relationships that inform patient-specific therapeutic interventions. Using reliability-centric ensemble ML approaches like random forest and XG Boost, alongside DL architectures including feedforward neural networks, the approach uncovers non-linear patterns and latent associations in patient behavior and treatment outcomes. To address the opacity often associated with complex models, XAI methods such as SHAP values and LIME are deployed to render outputs interpretable, thus aligning with medical ethics and regulatory standards. The results and Explainable AI methods, such as SHAP and LIME that are employed to interpret complex model predictions and ensure transparency for stakeholders, demonstrate how the insights can be translated into evidence-based pharmaceutical marketing mix modeling, supporting targeted interventions, efficient market segmentation, and personalized patient engagement strategies. The proposed approach bridges the gap between clinical research and marketing strategy, offering a data-driven pathway to enhance decision-making in the pharmaceutical sector while maintaining patient-centered practices. In this study, multiple machine learning and deep learning models, including support vector machine (SVM), random forest, XGBoost, logistic regression, feed forward neural network (FFNN), and multi-layer perceptron (MLP), were systematically evaluated for predicting cardiovascular disease outcomes using the Framingham Heart Study dataset. The SVM model demonstrated superior performance, achieving the highest test accuracy (96.65%), precision (96.55%), recall (87.50%), F1 score (91.80%), and ROC AUC (99.00%), outperforming all other baseline and deep learning models. In contrast, the FFNN and MLP models exhibited moderate performance, with final test accuracies of approximately 79.88% and 95.98%, respectively. The ensemble base learners, including XGBoost, random forest, and logistic regression, achieved lower accuracies (ranging from 71% to 79%) and reduced recall rates, indicating limitations in correctly identifying high-risk cases. Interpretability through LIME further validated the SVM’s robust decision boundaries and its alignment with clinically significant risk factors. Overall, the comparative results establish the SVM as the most reliable and generalizable predictive model for early cardiovascular risk stratification in the Framingham cohort.</p>