The medical sector has advanced in an imposing way, and are coming up with lifesaving models and wearable devices for disease predictions and patient monitoring. The prediction models and wearable devices will lead to immense amount of data collection leading to the dimensionality issues, overfitting and inaccurate results. From the pool of data that we use for our prediction model, we should be able to identify the required information and parameters which gives a positive contribution to the decision making model. Every dataset with higher number of parameters and high dimensionality will tend to the problems of overfitting. Here, we have a dataset of demented and non-demented patients with five conventional features and other physical parameters. Along with these parameters, we are adding three new prediction parameters like glyhb, BMI and Cholesterol, for proving the association of Diabetics and Dementia. After the addition of these parameters, the dataset will have thirty parameters, and dimensionality reduction is done to avoid the condition of overfitting. The work uses Principal Component Analysis(PCA)for reducing the dimensionality, t-SNE for visualization and K means clustering is used to cluster the target variable. The cluster mean of each variable is used to understand the performance of each variable in each cluster. Later, a basic feature ranking method is also implemented which can be further used for the prediction model. The performance metric used in this research work is Silhouette score, Inertia and Inter-Cluster Distance map.