PART I:
1.
2.
3.
The age, maximum heart rate (thalach), ST depression (oldpeak), and
cholesterol levels (chol) are the key variables that most clearly separate the
clusters. Age and thalach particularly offer strong separation between clusters,
with older patients typically clustering in C1, and younger patients in C2.
4. Cluster 1 (C1): Older patients (mean ≈ 59 years), lower heart rate, higher
cholesterol, and higher ST depression, which suggests more heart strain.
Cluster 2 (C2): Younger patients (mean ≈ 51 years), higher heart rate, lower
cholesterol, and low ST depression, indicating fewer signs of heart strain.
5. The silhouette scores for k-values suggest that k=2 offers the best clustering,
clearly separating the two groups. Medical insight can help prioritize healthcare
resources to monitor high-risk patients (C1) and ensure preventive care for lowrisk patients (C2). From a business perspective, clustering helps tailor patient
care and efficiently allocate resources.
PART II:
1.
The dataset was split using the Data Sampler widget with 70% for training and 30% for
testing, ensuring we have a good training set and a holdout set for evaluation.
2.
3.
Based on the Test & Score results, here are the evaluation metrics for the Logistic
Regression model:
Accuracy (CA): 0.838
Precision: 0.841
Recall: 0.838
F1-Score: 0.838
These results show a reasonable performance but suggest Random Forest may provide
better predictions.
4.
The features that play the largest role in predicting heart disease are thal (thalassemia),
exang (exercise induced angina), and cp (chest pain type), which are all indicative of
symptoms or risk factors related to cardiovascular health.
5.
Logistic Regression: Easier to interpret, but lower accuracy (83.8%) and AUC (0.917).
Random Forest: Higher accuracy (94.2%) and AUC (0.982), but less interpretable.
Random Forest performs better due to its ability to model non-linear relationships,
making it more suitable for complex data. However, it sacrifices interpretability
compared to Logistic Regression.
Random Forest offers higher accuracy and could be used for risk stratification, allowing
doctors to identify high-risk patients for early interventions.