Dementia, a disorder of progressive nature causes deterioration in cognitive function. Patients with dementia suffer from recalling through memory, language, comprehension, orientation, and judgment. Predicting dementia diagnosis is a critical task in healthcare, facilitating early detection and intervention for better patient outcomes. Feature importance analysis plays a pivotal role in understanding the factors influencing dementia prediction models. This paper proposes a framework to predict dementia using the open dataset collected from Kaggle. This research applies data imputation, pre-processing and transformation to generate pertinent data for model training. Furthermore, this research proposes a predictive dementia model exploring three well-known machine learning classifiers such as Random Forest, Decision Tree, and XGBoost to gain insights into the relative contributions of the most influential predictors. This proposed framework presents the optimal accuracy score with feature importance using SHAP analysis. Random Forest is identified to be the best algorithm with 94.74% accuracy. This study aims to guide future research directives in dementia diagnosis and to improve early detection and intervention strategies, ultimately enhancing patient care and outcomes.