Developing an approach to estimate the number of road traffic crashes at the national level using media-reported data.

The study presents a promising method for estimating road traffic crashes using media-reported data, achieving an R² of 0.93 with the artificial neural network model.

Summary and findings

This study developed a method to estimate the number of road traffic crashes using media-reported data in China. The approach utilized data from 379 media outlets and employed machine learning models to improve estimates compared to official statistics. The artificial neural network model achieved an R² of 0.93 for training data.

Abstract

<h4>Background</h4>Road traffic injury is a significant global health challenge, and timely available data are significant to monitor this trend. We aimed to develop a cost-effective approach in resource-limited settings to estimate the number of road traffic crashes at the national level by utilizing media-reported data.<h4>Methods</h4>Media-reported data about road traffic crashes were extracted from the Automated Road Traffic Crash Data Platform (ARTCDP) and augmented based on the available police reports with limited free-access. Besides, crash data were approximated according to the national disease surveillance point (DSP). We then fitted four common machine learning models (linear regression, artificial neural network, support vector machine, classification and regression tree) with six predictors to determine the best predictive model for road traffic crashes in China and correct underestimation of police-reported data.<h4>Results</h4>Of the 50,850 media outlets indexed by the ARTCDP, 379 media outlets reporting road traffic crash news quarterly were determined as the most reliable media-reported data sources. Of the four machine learning methods, artificial neural network performed best, yielding an <i>R</i> ² of 0.93 for training data, 0.92 for validation data, and 0.88 for testing. The number of road traffic crashes estimated by the approach closely matched actual trends in national number of crashes from official statistics.<h4>Conclusions</h4>Our approach based on ARTCDP-collected media outlets demonstrated excellent predictive performance and has potential to be used for estimating national road traffic crash statistics in resource-limited locations where official statistics are absent, not freely accessible, not reliable, or not yet released.

Key findings

• Not reported in abstract.
• Not reported in abstract.
• Not reported in abstract.

Limitations

• Not reported in abstract.
• Not reported in abstract.
• Not reported in abstract.

Elsewhere in the GHRP-6 corpus