Water Potability Prediction Project

Access to safe drinking water is a fundamental requirement for public health and a major factor in preventing waterborne diseases. This project implements a professional machine learning pipeline to analyze water quality parameters and predict whether a water source is potable (safe to drink) or non potable.

Project Overview

The core objective of this system is to automate the diagnostic process of water quality assessment. By utilizing chemical properties and physical measurements, the system provides a rapid classification tool that can support environmental health decision making.

Water Quality Parameters

The model evaluates nine critical factors found in the dataset to determine safety:

ph: The acid base balance of the water. Most potable water falls between 6.5 and 8.5.

Hardness: The concentration of calcium and magnesium salts, which affects how water interacts with pipes and soap.

Solids: The total dissolved solids (TDS) indicating the mineral content in the water.

Chloramines: The concentration of chlorine and ammonia used for disinfection.

Sulfate: Naturally occurring minerals; high levels can affect the taste and safety of the water.

Conductivity: A measure of the ability of water to pass an electrical current, indicating the presence of dissolved salts.

Organic Carbon: The total amount of carbon found in organic compounds, often used as a measure of water purity.

Trihalomethanes: Chemicals that can result from water treatment processes.

Turbidity: The measure of light scattering properties of the water, which indicates the presence of suspended particles.

Detailed Project Structure

The repository is organized following MLOps best practices to ensure that data, code, and model versions are strictly managed.

mlops-water-potability/
├── .dvc/                   # DVC configuration and cache (Internal)
├── .github/                # GitHub Actions for CI/CD (Optional)
├── data/                   # Data directory (Ignored by Git, tracked by DVC)
│   ├── raw/                # Original, immutable data (train.csv, test.csv)
│   └── processed/          # Data after cleaning/feature engineering
├── models/                 # Saved model artifacts (.pkl files)
├── reports/                # Metrics and plots (metrics.json)
├── src/                    # Source code directory
│   ├── __init__.py         # Makes 'src' a Python package
│   ├── data/               # Scripts related to data handling
│   │   ├── __init__.py
│   │   ├── data_collection.py
│   │   └── data_prep.py
│   └── model/              # Scripts related to ML logic
│       ├── __init__.py
│       ├── model_building.py
│       └── model_eval.py
├── .gitignore              # Tells Git what to ignore (data, .venv, models)
├── dvc.lock                # DVC state file (Crucial: Keep in Git)
├── dvc.yaml                # DVC pipeline definition
├── params.yaml             # Hyperparameters and configurations
├── requirements.txt        # Python dependencies
└── README.md               # Project overview and instructions

Installation and Setup

This project is optimized for Ubuntu. To ensure a clean environment, follow these steps:

- Create a virtual environment to isolate the project dependencies:

    python3 -m venv .venv

- Activate the virtual environment:

    source .venv/bin/activate

- Install the required libraries (Pandas, Scikit Learn, FastAPI, and DVC):

    pip install -r requirements.txt

Running the Pipeline

This project uses Data Version Control (DVC) to automate the workflow. Instead of running each script manually, you can execute the entire pipeline with a single command:

dvc repro

This command checks which parts of your code or data have changed and only runs the necessary steps to update your model and metrics. Prediction Service

The project includes a web service that allows you to interact with the model through an API.

Launch the service:

    uvicorn src.main:app --host 0.0.0.0 --port $PORT

Access the documentation:
Once the server is running, you can visit http://127.0.0.1:8000/docs in your browser to see the interactive API dashboard.

Submit a test: Use the test script to send a sample of water data to the server:

    python3 src/test.py

Model Evaluation

The model's performance is measured using several medical and statistical benchmarks:

Accuracy: Overall correctness of the model.

Precision: The ability to avoid false positives (labeling unsafe water as safe).

Recall: The ability to identify all truly potable water sources.

F1 Score: The balance between precision and recall.

These values are updated automatically in the metrics.json file every time the model is retrained.

License

This project is licensed under the MIT License. You are free to use, modify, and distribute this software for educational or professional purposes. See the LICENSE file for more details.