Do Traffic Congestion Indicators Improve PM2.5 Forecasting? Evidence from Machine Learning Models in Kampala

Abstract Air pollution remains a major environmental and public health challenge in Kampala, Uganda, where PM 2.5 concentrations frequently exceed World Health Organization (WHO) guidelines values. Although traffic emissions are widely recognized as contributors to urban air pollution, limited empirical evidence exists on whether traffic-congestion indicators improve PM 2.5 forecasting beyond information provided by meteorological and temporal variables. This study evaluated the predictive contribution of traffic-congestion indicators to PM 2.5 using machine learning models. PM 2.5 , temperature, and humidity observations were obtained from the AirQo monitoring network, while traffic data were collected along the Jinja Road transport using the Google Routes API. After preprocessing and temporal synchronization, a dataset consisting of 696 hourly observations was used to implement Random Forest and HistGradientBoosting forecasting models. Environmental baseline models were compared with traffic-enhanced models incorporating mean speed, travel duration, and Traffic Performance Index (TPI). Model performance was evaluated using Mean Absolute Error (MAE), Root Mean Squared Error (RMSE), and the coefficient of determination (R 2 ). Results showed that the best environmental baseline model, HistGradientBoosting, achieved superior performance (MAE = 15.28, RMSE = 21.40, R 2 = 0.5752) compared with the best traffic-enhanced model (MAE = 23.16, RMSE = 31.70, R 2 = 0.4753). Correlation and feature-importance analysis further indicated that traffic-congestion indicators exhibited weak relationships with PM2.5 concentrations and contributed substantially to less predictive information than temporal and meteorological variables. An additional analysis incorporating lagged PM 2.5 variables achieved the highest explanatory performance (R 2 = 0.6245), highlighting the importance of historical pollution measurements in forecasting future PM 2.5 concentrations. The findings suggest that, within the study context, corridor-level congestion indicators provided limited incremental forecasting value, while temporal patterns, meteorological conditions, and historical PM 2.5 measurements were more influential predictors. The study contributes empirical evidence for evaluating predictor relevance in machine-learning based air-quality forecasting and provides insights for the design of efficient forecasting systems in rapidly urbanizing cities.

Authors

Institutions

Publication Details

Journal
Research Square
Published
2026-11-16
DOI
https://doi.org/10.21203/rs.3.rs-11720518/v1
Primary Topic
Air Quality Monitoring and Forecasting
Type
preprint
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
preprint

Do Traffic Congestion Indicators Improve PM2.5 Forecasting? Evidence from Machine Learning Models in Kampala

Hedwig Orieba, Deborah Mackenzi
2 links
Research Square
Air Quality Monitoring and Forecasting
preprint

Do Traffic Congestion Indicators Improve PM2.5 Forecasting? Evidence from Machine Learning Models in Kampala

Hedwig Orieba, Deborah Mackenzi
preprint en

Abstract

Abstract Air pollution remains a major environmental and public health challenge in Kampala, Uganda, where PM 2.5 concentrations frequently exceed World Health Organization (WHO) guidelines values. Although traffic emissions are widely recognized as contributors to urban air pollution, limited empirical evidence exists on whether traffic-congestion indicators improve PM 2.5 forecasting beyond information provided by meteorological and temporal variables. This study evaluated the predictive contribution of traffic-congestion indicators to PM 2.5 using machine learning models. PM 2.5 , temperature, and humidity observations were obtained from the AirQo monitoring network, while traffic data were collected along the Jinja Road transport using the Google Routes API. After preprocessing and temporal synchronization, a dataset consisting of 696 hourly observations was used to implement Random Forest and HistGradientBoosting forecasting models. Environmental baseline models were compared with traffic-enhanced models incorporating mean speed, travel duration, and Traffic Performance Index (TPI). Model performance was evaluated using Mean Absolute Error (MAE), Root Mean Squared Error (RMSE), and the coefficient of determination (R 2 ). Results showed that the best environmental baseline model, HistGradientBoosting, achieved superior performance (MAE = 15.28, RMSE = 21.40, R 2 = 0.5752) compared with the best traffic-enhanced model (MAE = 23.16, RMSE = 31.70, R 2 = 0.4753). Correlation and feature-importance analysis further indicated that traffic-congestion indicators exhibited weak relationships with PM2.5 concentrations and contributed substantially to less predictive information than temporal and meteorological variables. An additional analysis incorporating lagged PM 2.5 variables achieved the highest explanatory performance (R 2 = 0.6245), highlighting the importance of historical pollution measurements in forecasting future PM 2.5 concentrations. The findings suggest that, within the study context, corridor-level congestion indicators provided limited incremental forecasting value, while temporal patterns, meteorological conditions, and historical PM 2.5 measurements were more influential predictors. The study contributes empirical evidence for evaluating predictor relevance in machine-learning based air-quality forecasting and provides insights for the design of efficient forecasting systems in rapidly urbanizing cities.

Research Square
Victoria University (AU)
Sustainable cities and communities, Industry, innovation and infrastructure, Good health and well-being
Air Quality Monitoring and Forecasting
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon