🩺 Multiclass Diabetes Classification

📌 Overview

This project aims to predict the type of diabetes (Type 1, Type 2, or Non-diabetic) based on patients' medical attributes.
The dataset used is a cleaned and balanced subset of the Multiclass Diabetes Dataset from Kaggle/Mendeley, which has been processed to remove duplicates and partially balance the classes.

---

📊 Dataset

• Source: Multiclass Diabetes Dataset – Kaggle / Mendeley Data
• Original records: 1,000
• After removing duplicates: 826 unique samples
  • Diabetic: 690
  • Non-diabetic: 96
  • Pre-diabetic: 40
• Final subset used in this project:
  • Diabetic: 128
  • Non-diabetic: 96
  • Pre-diabetic: 40

Target variable (Class):

• 0 → Non-diabetic
• 1 → Pre-diabetic (Type 1)
• 2 → Diabetic (Type 2)

---

🛠 Workflow

1. Exploratory Data Analysis (EDA)
   • Dataset shape, data types, summary statistics
   • Class distribution visualization
   • Correlation heatmap for feature relationships
2. Preprocessing
   • Splitting features and target column
   • Train-test split with stratification
   • Feature scaling using StandardScaler
3. Handling Class Imbalance
   • Applied SMOTE with sampling_strategy={1: 96} to oversample Class 1 to match Class 0 size
4. Model Training & Evaluation
   • Models used:
     • Logistic Regression
     • RandomForestClassifier
     • XGBClassifier
     • K-Nearest Neighbors (KNN)
   • Evaluation metrics:
     • Accuracy
     • Precision, Recall, F1-score
     • Confusion Matrix
   • Accuracy comparison plotted for all models

---

📈 Results

Model	Accuracy	Key Notes
Logistic Regression	0.89	Class 1 performance improved after SMOTE
RandomForestClassifier	0.98	Best overall performance, high precision & recall for all classes
XGBClassifier	0.96	Close to RandomForest
KNN	~0.79	Lower performance compared to others

Main observations:

• SMOTE improved minority class (Class 1) performance significantly.
• RandomForestClassifier achieved the best accuracy (98%) and overall metrics.