🩺 PCOS Prediction Model — R Project

A machine learning project that predicts Polycystic Ovary Syndrome (PCOS) using clinical and lifestyle data. Three models are trained and compared: Logistic Regression, Random Forest, and XGBoost.

---

📁 Project Structure

PCOS-Prediction-R/
│
├── data/
│   └── pcos_dataset.csv          # Raw dataset (input)
│
├── output/
│   ├── predictions.csv           # Test set predictions
│   ├── pcos_prediction_model.rds # Saved Random Forest model
│   ├── pcos_clean.rds            # Cleaned dataset
│   ├── train_data.rds            # Training split
│   ├── test_data.rds             # Test split
│   └── cor_matrix.rds            # Correlation matrix
│
├── screenshots/
│   └── dashboard.png             # Project dashboard preview
│
├── scripts/
│   ├── data_cleaning.R           # Step 1 — Load & clean data
│   ├── eda.R                     # Step 2 — Exploratory analysis (14 plots)
│   ├── model_training.R          # Step 3 — Train, compare & evaluate models
│   └── prediction.R              # Step 4 — Predict new patients
│
├── PCOS_Prediction.R             # Master script (runs all steps)
├── README.md                     # This file
└── requirements.txt              # R package dependencies

---

🚀 Quick Start

1. Install R and VS Code

• Download R: https://cran.r-project.org/
• Download VS Code: https://code.visualstudio.com/
• Install R extension in VS Code: REditorSupport.r

2. Install required packages

Open R console and run:

install.packages(c(
  "tidyverse", "caret", "randomForest",
  "xgboost", "pROC", "corrplot", "pdp"
), dependencies = TRUE)

3. Set your data path

Open scripts/data_cleaning.R and update line 10:

DATA_PATH <- "E:/projects/PCOS-Prediction-R/data/pcos_dataset.csv"

4. Run scripts in order

source("scripts/data_cleaning.R")   # Clean data
source("scripts/eda.R")             # Explore data
source("scripts/model_training.R")  # Train models
source("scripts/prediction.R")      # Make predictions

Or run the master script:

source("PCOS_Prediction.R")

5. Run from terminal (after adding R to PATH)

& "C:\Program Files\R\R-4.5.3\bin\Rscript.exe" "PCOS_Prediction.R"

---

📊 Dataset

Property	Details
Source	Kaggle — PCOS Dataset
Rows	~541 patients
Target	PCOS_Y_N (Yes / No)
Features	41 clinical + lifestyle variables

Key Features Used

• Demographics — Age, BMI, Blood Group
• Symptoms — Weight gain, Hair growth, Skin darkening, Hair loss, Pimples
• Lifestyle — Fast food consumption, Regular exercise
• Clinical — Cycle regularity, Pregnancy history, Marriage duration
• Hormonal — FSH, LH, TSH, AMH levels

---

🤖 Models

Model	Algorithm	Cross-Validation
Logistic Regression	glm (binomial)	10-fold CV
Random Forest	randomForest	5-fold CV
XGBoost	xgbTree	5-fold CV

Best model selection is based on AUC (ROC) score.

---

📈 Results (Random Forest — Best Model)

Metric	Score
Accuracy	~88%
Sensitivity	~85%
Specificity	~90%
AUC	~0.94

Results may vary slightly depending on your dataset version and random seed.

---

🔮 Making Predictions

# Load the saved model
model      <- readRDS("pcos_prediction_model.rds")
train_data <- readRDS("train_data.rds")

# Create a patient record
new_patient <- data.frame(
  Age..yrs.        = 28,
  BMI              = 27.5,
  Cycle.R.I.       = factor("I", levels = levels(train_data$Cycle.R.I.)),
  Weight.gain.Y.N  = factor("1", levels = levels(train_data$Weight.gain.Y.N)),
  hair.growth.Y.N  = factor("1", levels = levels(train_data$hair.growth.Y.N))
  # ... add more columns as needed
)

# Predict
prediction  <- predict(model, newdata = new_patient)
probability <- predict(model, newdata = new_patient, type = "prob")

cat("Diagnosis   :", as.character(prediction), "\n")
cat("Probability :", round(probability$PCOS, 4), "\n")

---

📦 Output Files

File	Description
pcos_prediction_model.rds	Trained Random Forest model
output/predictions.csv	Test set actual vs predicted results
pcos_clean.rds	Cleaned and preprocessed dataset
train_data.rds	70% training split
test_data.rds	30% test split

---

⚠️ Disclaimer

This model is built for educational and research purposes only.
It is not a medical diagnostic tool.
Always consult a qualified healthcare professional for medical advice.

---

👩‍💻 Tech Stack