Cluster Analysis in R

🎯 Problem Statement

This project applies unsupervised cluster analysis techniques to the widely used Framingham Heart Study dataset using the R programming language. The core goal is to identify distinct groups within the patient population that exhibit similar cardiovascular health risk factors. The analysis provides a foundation for risk stratification, targeted public health intervention programs, and optimizing clinical resource allocation for preventative care.

Tools Used: R (Programming Language), dist, hclust, factoextra, NbClust (R Libraries)

---

🗃️ Dataset Overview

The dataset used is the Framingham Heart Study, which contains multiple physiological and demographic variables. While originally a medical study, the data is used here to demonstrate the ability to segment high-dimensional data based on multiple features simultaneously.

• Key Action: Data preprocessing involved handling missing values (na.omit) and standardizing features to ensure all variables contributed equally to the distance calculation.

Dataset Source: Kaggle Dataset

---

🔧 Methodology & R Code Highlights

-Preprocessed data (handled missing values, standardized features). -Computed Euclidean distance matrix. -Performed hierarchical clustering (Ward’s method) and k-means clustering. -Visualized cluster structures using dendrograms and scatter plots.

1. Data Preprocessing

Data was loaded and cleaned to ensure robust model performance.

library(readr)
framingham <- read.csv("Framingham.csv")
framingham_1 <- na.omit(framingham)

2. Determining Optimal Cluster Count

The NbClust package was used to statistically determine the optimal number of clusters (k), minimizing within-cluster variance while maximizing between-cluster variance.

library(NbClust)
nb.fit <- NbClust(framingham_1, distance = "euclidean", min.nc=2, max.nc=10, method = "average")

Optimal Number of Clusters

3. Hierarchical Clustering (H-Clust)

H-Clust was performed using Ward's minimum variance method (ward.D) to generate the cluster hierarchy, which provides a visual structure of the data's segmentation.

DIST <- dist(framingham_1, method = "euclidean")
H_TREE <- hclust(DIST, method = "ward.D")
fviz_dend(H_TREE, cex = 0.5)

4.Cluster Visualization

Clusters<-cutree(H_TREE , k=2)
fviz_cluster(list(H_TREE, data = framingham_1, cluster = Clusters))

Clusters Visualization

5. Final Cluster VisualizationThe optimal number of clusters ($k=2$) was cut from the hierarchy and visualized, clearly separating the two primary groups based on their feature characteristics.

# Cut the tree into 2 groups and visualize
Clusters <- cutree(H_TREE, k=2)
fviz_cluster(list(H_TREE, data = framingham_1, cluster = Clusters))

Dendrogram

💡Key Insights

• Clear Segmentation: Both Hierarchical Clustering and k-means validated that two distinct segments are present in the data, as evidenced by the NbClust result.

• Actionable Grouping: The identified clusters exhibit clear statistical differences in their means across key variables (e.g., age, blood pressure, etc.), allowing for the definition of distinct customer profiles (e.g., "High-Risk Segment" vs. "Low-Risk Segment").

• Methodological Proof: The project demonstrates proficiency in standardizing data, measuring distance, and validating cluster stability—essential skills for any segmentation task.