Machine Learning

🚀 Inspiration Air pollution is a critical global issue affecting millions of lives. Predicting air quality can help in health monitoring, environmental safety, and policy-making. This project was inspired by the need to create a machine learning model capable of accurately predicting Relative Humidity (RH) based on air pollutants and meteorological data.

🌍 What it does This project: ✅ Preprocesses air quality data by handling missing values, formatting timestamps, and scaling features. ✅ Identifies correlations and removes redundant features to improve model efficiency. ✅ Trains and optimizes a Stacking Regressor using Random Forest and Gradient Boosting models. ✅ Predicts Relative Humidity (RH) based on various environmental factors. ✅ Visualizes results with error distributions and actual vs. predicted values.

🛠 How we built it Data Processing: Cleaned, transformed, and normalized the dataset. Feature Engineering: Extracted Month, Day, and Hour from timestamps. Model Selection: Used Stacking Regression with Random Forest & Gradient Boosting. Hyperparameter Tuning: Applied Optuna to optimize model parameters efficiently. Evaluation: Used R² Score, MAE, RMSE to measure model performance. ⚡ Challenges we ran into 🔹 Handling missing values and outliers in the dataset. 🔹 Reducing training time while maintaining high accuracy. 🔹 Tuning hyperparameters effectively without overfitting.

🏆 Accomplishments that we're proud of ✔ Successfully optimized the model using Optuna for hyperparameter tuning. ✔ Achieved a higher R² score with reduced model training time. ✔ Built an efficient air quality prediction system that can be further expanded.

📚 What we learned 🔸 Importance of feature selection in reducing model complexity. 🔸 The power of Stacking Regression for improved predictions. 🔸 How hyperparameter tuning significantly enhances performance.

🚀 What's next for Machine Learning in Air Quality Prediction? 🔹 Incorporating real-time air quality data from APIs. 🔹 Expanding predictions to include pollutants like CO, NO2, and SO2. 🔹 Deploying the model as a web application for public use.

Built With

• dataset
• datetime
• gradientboosting
• kaggle
• machine-learning
• matplotlib
• numpy
• optuna
• pandas
• python
• randomforest
• scikit-learn
• seaborn
• stacking
• standardscalar