Stroke Risk Classification | Using Random Forest Algorithm and SMOTE Mehthod (to overcome dataset imbalance)

Overview

This project focuses on predicting stroke risk using machine learning classification techniques based on demographic, lifestyle, and medical information. The objective is to identify individuals who may have a higher likelihood of experiencing a stroke and evaluate the effectiveness of the Random Forest algorithm in handling healthcare classification problems.

---

Problem Statement

Stroke is one of the leading causes of death and long-term disability worldwide. Early detection of high-risk individuals can assist healthcare providers in implementing preventive measures and improving patient outcomes.

This project aims to build a classification model capable of predicting stroke occurrence using patient health data.

---

Dataset

Source: Kaggle Healthcare Stroke Dataset (Link: https://www.kaggle.com/datasets/ranaghulamnabi/stroke-risk-dataset)

The dataset contains demographic and medical information such as:

• Gender
• Age
• Hypertension
• Heart Disease
• Marital Status
• Work Type
• Residence Type
• Average Glucose Level
• BMI
• Smoking Status

Target Variable

| Value | Description | | 0 | No Stroke | | 1 | Stroke |

---

Project Workflow

1. Data Understanding

• Load dataset
• Examine data structure
• Identify missing values
• Check duplicate records

2. Data Preprocessing

• Remove unnecessary columns (id)
• Handle missing values
• Encode categorical variables using Label Encoding

3. Exploratory Data Analysis (EDA)

Visualizations performed:

• Stroke Distribution
• Gender Distribution
• Age Distribution
• BMI Distribution
• Smoking Status Distribution

---

4. Data Splitting

Dataset split into:

• Training Set (80%)
• Testing Set (20%)

---

5. Handling Imbalanced Data

The dataset is highly imbalanced.

To address this issue:

• SMOTE (Synthetic Minority Oversampling Technique) was applied to the training data.

Benefits:

• Better representation of minority class
• Improved model learning
• Reduced bias toward majority class

---

6. Model Development

Algorithm used:

Random Forest Classifier

Parameters:

RandomForestClassifier(
    n_estimators=200,
    class_weight='balanced',
    random_state=42
)

Reason for choosing Random Forest:

• Robust against overfitting
• Handles non-linear relationships
• Works well on tabular healthcare data
• Provides feature importance analysis

---

Model Evaluation

Evaluation metrics:

• Accuracy
• Precision
• Recall
• F1-Score
• Confusion Matrix
• Feature Importance Analysis
• ROC and AUC Curve

Stroke Distribution

---

Confusion Matrix

---

Feature Importance Analysis

Feature importance was extracted from the trained Random Forest model to identify the most influential factors contributing to stroke prediction.

Feature Importance Visualization

---

In Used

• Python
• Pandas
• NumPy
• Matplotlib
• Seaborn
• Scikit-Learn
• Imbalanced-Learn (SMOTE)
• Jupyter Notebook or Vs Code (ipynb)

---

Project Structure

stroke-risk-classification/
│
├── Dataset Kaggle Stroke.csv
│   
│
├── STROKE_Random_Forest.ipynb
│   
│
├── images/
│   ├── stroke_distribution.png
│   ├── confusion_matrix.png
│   ├── feature_importance.png
│   ├── before_smote.png
│   └── after_smote.png
│
├── README.md
│
└── requirements.txt

---

Key Findings

• The original dataset contains class imbalance between stroke and non-stroke cases.
• SMOTE significantly improved minority class representation.
• Random Forest successfully learned patterns from healthcare attributes (91% Accuration).
• Feature importance analysis reveals the most influential factors associated with stroke occurrence.

---

Future Improvements

• Hyperparameter Tuning
• Cross Validation
• XGBoost Comparison
• LightGBM Comparison
• Model Deployment using Streamlit
• Use Ensemble Learning
• SMOTE + ENN

---