Traffic NOx Emissions Regression Model

ENGR-125 | Machine Learning Regression

Project Overview

This project focuses on predicting Nitrogen Oxide (NOx) pollution using traffic and environmental variables (humidity, temperature, windspeed, car/truck counts, and wind direction). NOx exposure is strongly linked to respiratory issues and can worsen asthma symptoms over time, making accurate prediction valuable for public health and environmental engineering.

Dataset Preview

') }}" "Traffic dataset shown in Excel"

The dataset was cleaned by removing date/time columns, inspecting missing values, and splitting the data into training and testing sets. PM2.5 was excluded due to its similarity with NOx, which created issues during training and prediction.

Model Development

• Data cleaning and preprocessing
• Train/test split and feature/label separation
• Normalization layer for numeric inputs
• Linear regression (single-variable and multi-variable)
• Deep Neural Network regression (TensorFlow / Keras)
• Prediction and error analysis

Results

}}" "Model results: predicted vs actual / training performance"

The trained model showed a noticeable relationship between humidity levels and NOx output. Predicted vs. actual plots showed a reasonable fit, with some variance remaining in edge cases. Loss curves indicated the model learned meaningful patterns without extreme overfitting.

Challenges

• Selecting the best variables for stable learning
• Removing date/time to avoid formatting and training errors
• Dropping PM2.5 because it interfered with prediction performance

Real-World Application

This model can help environmental engineers and scientists forecast NOx pollution trends and support decisions related to traffic planning, emissions reduction, and public health protection.

Skills Demonstrated

• Python (Pandas, NumPy)
• Data visualization (Matplotlib, Seaborn)
• TensorFlow / Keras regression modeling
• Data preprocessing and feature engineering
• Engineering focused analysis and reporting