🌍 Air Quality Index (AQI) Prediction

📌 Project Overview

Air pollution is a critical environmental issue, and predicting air quality is essential for public health and policymaking.
This project leverages Machine Learning (ML) techniques to forecast Air Quality Index (AQI) based on key environmental pollutants.
The models implemented include Linear Regression, Random Forest, and XGBoost, with XGBoost emerging as the best-performing model.

---

📂 Project Resources

🔹 Dataset (CSV File): Download city_day.csv
🔹 Dataset (Kaggle Link): View on Kaggle
🔹 Project Code (.ipynb): View Jupyter Notebook
🔹 Presentation (PPTX File): Download Project Report

---

📊 Dataset Overview

The dataset used for this project contains 29,531 records and 16 features, including:

• Pollutants: PM2.5, PM10, NO₂, SO₂, CO, O₃, Benzene, etc.
• Date & City: Identifying the location and time of recording.
• AQI (Target Variable): Measures pollution severity and categorizes it into buckets (Good, Moderate, Poor, etc.).

---

🛠 Data Preprocessing

✔️ Handling Missing Values:

• Removed rows with missing AQI values.
• Imputed missing pollutant values using mean imputation.

✔️ Feature Scaling:

• Used StandardScaler to standardize numerical features (zero mean, unit variance).

✔️ Feature Selection:

• Analyzed feature correlation using a heatmap to determine the most influential pollutants.
• PM2.5 (0.65) and CO (0.68) showed the highest correlation with AQI.

---

🤖 Models Used

The project explores different regression models to predict AQI:

1️⃣ Linear Regression

🔹 Simple and interpretable but struggles with complex patterns.

2️⃣ Random Forest Regressor

🔹 Uses multiple decision trees for better accuracy and handles non-linearity.

3️⃣ XGBoost Regressor (🏆 Best Model)

✅ More accurate, faster, and better at handling missing data than Random Forest.
✅ Captures complex relationships effectively with boosting techniques.
✅ Reduces overfitting and performs well on large datasets.

---

📈 Model Evaluation

To measure performance, we used the following metrics:

📌 Mean Absolute Error (MAE) – Lower values indicate better accuracy.
📌 Root Mean Squared Error (RMSE) – Penalizes large prediction errors.
📌 R² Score (Coefficient of Determination) – Measures variance explained by the model.

📌 Model Comparison:

Model	MAE	RMSE	R² Score
Linear Regression	High	High	Low
Random Forest	Moderate	Moderate	Moderate
XGBoost 🏆	Low	Low	High

---

🔧 Hyperparameter Tuning

To improve the XGBoost model, we optimized the following parameters using GridSearchCV:

• n_estimators: [50, 100, 200]
• learning_rate: [0.01, 0.1, 0.2]
• max_depth: [3, 5, 7]

✅ Best Configuration:
learning_rate = 0.2, max_depth = 5, n_estimators = 100

The final model was saved using joblib for future predictions.

---

📌 Results & Predictions

• The trained XGBoost Regressor effectively predicts AQI values for unseen data.
• A line plot was used to visualize Actual vs. Predicted AQI, confirming the model's accuracy.

---

🚀 Future Scope

🔹 Improve accuracy using Deep Learning (LSTMs, CNNs).
🔹 Integrate real-time AQI data & meteorological features.
🔹 Develop AI-powered dashboards & mobile apps for real-time AQI tracking.

---

🏆 Individual Contribution

This was a group project, but I contributed to the entire project, including:
✔️ Data Preprocessing & Cleaning
✔️ Model Implementation & Evaluation
✔️ Hyperparameter Tuning & Optimization
✔️ Model Interpretation & Documentation

---