🚰 Water Potability Prediction using Machine Learning

📌 Project Overview

This project predicts whether water is safe (potable) or unsafe for drinking, using physicochemical features.
The dataset consists of 3,276 water samples with 9 features (e.g., pH, hardness, solids, sulfate).

---

🔑 Problem

Clean drinking water is essential for health. Laboratory testing is reliable but costly and time-consuming.
This project aims to build a data-driven screening tool for water safety classification.

---

🧾 Dataset

• Source: (https://www.kaggle.com/datasets/adityakadiwal/water-potability)) (Water Potability dataset)
• Features:
  • ph
  • Hardness
  • Solids
  • Chloramines
  • Sulfate
  • Conductivity
  • Organic_carbon
  • Trihalomethanes
  • Turbidity
• Target:
  • Potability (1 = drinkable, 0 = not drinkable)

---

⚙️ Workflow / Pipeline

1. Data cleaning & handling missing values
2. Outlier detection & mitigation
3. Feature engineering (pH categories)
4. Train-test split
5. No leakage: scaling applied to train only
6. SMOTE: handle class imbalance on train set
7. Train models separately:
   • Logistic Regression
   • Support Vector Machine
   • Decision Tree
   • Random Forest
   • XGBoost
8. Model evaluation with accuracy, precision, recall, F1, ROC-AUC
9. Feature importance & SHAP interpretability

---

📊 Results

Model	Accuracy	Precision	Recall	F1 Score
Logistic Regression	0.50	0.40	0.56	0.47
SVM	0.62	0.52	0.44	0.48
Decision Tree	0.71	0.62	0.64	0.63
Random Forest	0.75	0.69	0.63	0.66
XGBoost	0.73	0.68	0.61	0.64

➡️ Random Forest achieved the best balance (F1 = 0.66, AUC ≈ 0.74).

---

🔍 Key Insights

• Safe water is more likely when pH is between 6.5–8.5 (WHO standard), but pH alone is not sufficient.
• Sulfate, Solids, and Conductivity were strong predictors in feature importance analysis.
• No single feature separates safe vs. unsafe water → ML models are necessary.
• Random Forest + XGBoost both performed strongly; RF slightly better on recall and F1.

---

📈 Visuals

• Confusion matrix
• ROC curve
• Feature importance (Random Forest)
• Model performance comparison chart

---

🏆 Conclusion

• Random Forest was the best performing model (F1 = 0.66).
• While this tool is not a replacement for laboratory testing, it provides a fast, low-cost, first-pass screening for water safety assessment. pip install -r requirements.txt jupyter notebook Water_Prediction.ipynb

🚀 How to Run

git clone https://github.com/sherryyy00/water-potability-ml.git
cd water-potability-ml