Task 3: Heart Disease Prediction

DevelopersHub Corporation — AI/ML Engineering Internship

Objective

Build a binary classification model to predict whether a patient is at risk of heart disease based on their health indicators.

Dataset

• Name: UCI Heart Disease Dataset (Cleveland)
• Source: Embedded directly inside the notebook — no internet or download required
• Size: 244 patients, 14 columns
• Target: 0 = No Disease, 1 = Heart Disease

Feature	Description
age	Age of the patient (years)
sex	Sex (1 = male, 0 = female)
cp	Chest pain type (0–3)
trestbps	Resting blood pressure (mm Hg)
chol	Serum cholesterol (mg/dl)
fbs	Fasting blood sugar > 120 mg/dl
restecg	Resting ECG results
thalach	Maximum heart rate achieved
exang	Exercise induced angina
oldpeak	ST depression induced by exercise
slope	Slope of peak exercise ST segment
ca	Number of major vessels (0–4)
thal	Thalassemia type
target	Heart Disease (1 = Disease, 0 = No Disease)

Tools & Libraries

• Python 3.10
• pandas, numpy
• scikit-learn (LogisticRegression, DecisionTreeClassifier)
• matplotlib, seaborn

What's Inside the Notebook

Step	Description
1. Import Libraries	All required libraries
2. Load Dataset	Loaded from embedded CSV — no internet needed
3. Feature Descriptions	Explanation of all 14 columns
4. Data Inspection	.info(), .describe(), missing values, class balance
5. EDA	Target distribution, age histogram, gender chart, box plots, heatmap
6. Preprocessing	Train/test split (80/20), StandardScaler
7. Model Training	Logistic Regression & Decision Tree
8. Evaluation	Accuracy, classification report, confusion matrix, ROC curve, feature importance
9. Insights	Key findings and conclusions

Model Performance

Model	Accuracy	AUC-ROC
Logistic Regression	~85%	~0.91
Decision Tree (depth=5)	~79%	~0.86

Key Results & Findings

• Logistic Regression outperforms Decision Tree in AUC-ROC — better suited for medical use
• thalach (max heart rate) is the strongest predictor of heart disease
• cp (chest pain type) and oldpeak (ST depression) are also top predictors
• Dataset is well balanced (~51% disease vs ~49% no disease) — no oversampling needed
• Both models significantly outperform a random classifier on the ROC curve

How to Run

pip install pandas numpy matplotlib seaborn scikit-learn jupyter
jupyter notebook Task3_Heart_Disease_Prediction.ipynb

Run all cells from top to bottom. No internet connection or dataset download required.