KNN-diabetes-classifier

This project uses the K-Nearest Neighbors (KNN) algorithm for two classification tasks:

• Diabetes Prediction using the Pima Indians dataset
• Wine Quality Classification using the Wine Quality dataset. It includes data preprocessing, normalization, model evaluation across various k-values, and a manual input interface for real-time predictions. Implemented in Python with scikit-learn and visualized using Matplotlib and Seaborn.

Problem Statement

Using sample dataset, apply the k-nearest neighbor classifier and analyse its performance with different values of k.

• Deliverable: Jupyter notebook/Excel file with all steps and brief documentation.

Dataset

Here we have used the same model on 2 different datasets

• Pima Indians Diabetes Dataset
• Wine Quality Dataset( Introductory Paper : Modeling wine preferences by data mining from physicochemical properties )

1. Source: Pima Indians Diabetes Database Description: Contains diagnostic measurements from female patients of Pima Indian heritage, including: Pregnancies Glucose BloodPressure SkinThickness Insulin BMI DiabetesPedigreeFunction Age

   Target: Outcome (0 = Not Diabetic, 1 = Diabetic)

2. Source : UC Irvine Machine Learning Repository The dataset used is winequality-red.csv. using this dataset model classifies red wine samples into Low, Medium, and High quality categories using the K-Nearest Neighbors (K-NN) algorithm

Diabetes Dataset Feature Interpretations

Pregnancies

Number of times the patient has been pregnant

• Typical Range: 0 to ~17
• Insight: Higher values may suggest increased risk, as gestational diabetes is a known factor.

Glucose

Plasma glucose concentration after 2 hours in an oral glucose tolerance test

• Typical Range:
  • Normal: 70–99 mg/dL
  • Prediabetes: 100–125 mg/dL
  • Diabetes: 126+ mg/dL
• Insight: High glucose is a strong indicator of diabetes.

BloodPressure

Diastolic blood pressure (mm Hg)

• Normal Range: ~60–80 mm Hg
• Insight: High or very low diastolic pressure may contribute to health risks.

SkinThickness

Triceps skinfold thickness (in mm)

• Typical Range: 10–50 mm
• Insight: Measures subcutaneous fat; correlates with BMI and obesity.

Insulin

2-Hour serum insulin (mu U/ml)

• Typical Range: 16–166 mu U/ml (can vary)
• Insight: Elevated or very low insulin levels can be indicators of insulin resistance.

BMI (Body Mass Index)

Weight (kg) / Height (m)^2

• Interpretation:
  • Underweight: < 18.5
  • Normal: 18.5–24.9
  • Overweight: 25–29.9
  • Obese: 30+
• Insight: Obesity is a major risk factor for type 2 diabetes.

DiabetesPedigreeFunction (Pedigree)

A function that scores likelihood of diabetes based on family history

• Typical Range: 0.0 – 2.5+
• Insight: Higher values mean a stronger genetic influence.

Age

Age in years

• Typical Range in Dataset: 21 – ~80
• Insight: Risk increases significantly with age, especially over 45.

Project Features

Data cleaning (replacing invalid zeros with median values)

Feature normalization using MinMaxScaler

Training and evaluating KNN for different values of k (1–20)

Accuracy visualization across k values

Confusion matrix and classification report

Manual input prediction with human-readable output (Diabetic or Not Diabetic)

Tech Stack

Python
Pandas, NumPy, Matplotlib, Seaborn
scikit-learn

How to Use

Clone the repository: git clone https://github.com/your-username/knn-diabetes-predictor.git cd knn-diabetes-predictor

Run the notebook: Open KNN_Diabetes_Analysis.ipynb in Jupyter or Google Colab and run all cells. Enter new patient data in the input cell to predict diabetes status.

Performance

1. Confusion Matrix

[[89 10]

[21 34]]

• Confusion matrix is for test-dataset

What this means:

                 Predicted: 0         Predicted: 1

Actual: 0 89 (True Negatives) 10 (False Positives)

Actual: 1 21 (False Negatives) 34 (True Positives)

True Negative (TN) = 89 → Correctly predicted class 0

False Positive (FP) = 10 → Incorrectly predicted class 1 when it was 0

False Negative (FN) = 21 → Incorrectly predicted class 0 when it was 1

Overall Metrics

Accuracy = 0.80 → 80% of total predictions were correct.

Macro avg: Averages precision, recall, and F1 equally across both classes, regardless of how many samples each has. Shows fairness across classes.

Weighted avg: Averages precision, recall, and F1 weighted by support (i.e., more weight to class 0 due to more samples). More reflective of overall model performance.

Sample Output

Enter the following values: Pregnancies: 2 Glucose: 130 BloodPressure: 70 SkinThickness: 28 Insulin: 88 BMI: 32.5 Pedigree: 0.45 Age: 36

Prediction Result: Not Diabetic

2 Wine Quality Dataset

Source : UC Irvine Machine Learning Repository The dataset used is winequality-red.csv. using this dataset model classifies red wine samples into Low, Medium, and High quality categories using the K-Nearest Neighbors (K-NN) algorithm

Dataset Overview

• Dataset: winequality-red.csv
• Samples: 1,599
• Features: 11 physicochemical inputs
• Target Variable: Wine quality (original score from 3 to 8)

Features Description

The dataset includes the following input features:

Feature Name	Description
Fixed Acidity	Tartaric acid content
Volatile Acidity	Acetic acid content
Citric Acid	Citric acid content
Residual Sugar	Remaining sugar after fermentation
Chlorides	Salt content
Free Sulfur Dioxide	Free SO₂ in wine
Total Sulfur Dioxide	Total SO₂ in wine
Density	Wine density
pH	Acidity level
Sulphates	Sulfate concentration
Alcohol	Alcohol content (%)

Target Variable Transformation

The quality score is transformed into 3 categories(Binning):

• Low (0): Scores 3 to 5
• Medium (1): Score 6
• High (2): Scores 7 to 8

This converts the problem into a multi-class classification task.

Preprocessing Steps

• Feature Scaling:
  • Applied both StandardScaler and MinMaxScaler
• Train-Test Split:
  • 80% training and 20% testing
  • Stratified to maintain class distribution

Model: K-Nearest Neighbors

• Hyperparameter Tuning:
  • Explored k values from 4 to 32
• Evaluation Metric:
  • Accuracy on the test set for each value of k

Best Performing k

• Best k: 8 with Accuracy: 0.6625

Confusion Matrix

Actual → / Predicted ↓	Low	Medium	High
Low	111	30	0
Medium	54	69	9
High	4	24	19

• Confusion matrix is for test-dataset

Classification Report

	precision	recall	f1-score	support
Low	0.66	0.79	0.72	141
Medium	0.56	0.52	0.54	132
High	0.68	0.40	0.51	47

Manual Prediction Example

Input: 7.3, 0.65, 0.0, 1.2, 0.065, 15.0, 21.0, 0.9946, 3.39, 0.47, 10.0

• Predicted Wine Quality: Medium