Thai University RankingsRESEARCH RADAR
Evidence of global relevance

Air Pollution-Related Mortality in Bangkok: A Time-Series Neural Network Analysis

A 2016-2020 Bangkok time-series analysis used RNN, LSTM and GRU models to relate six pollutants to daily mortality. A 23-day-lag LSTM was reported as the best model, while SHAP emphasised humidity, PM2.5 and ozone overall. Prediction does not equal causal attribution of deaths to pollution.

01

Key findings

  • The study reported a median of 36 premature deaths per day and selected a 23-day-lag LSTM. SHAP highlighted humidity, PM2.5 and O3 overall; PM10, PM2.5 and CO for older adults; and PM2.5, NO2 and PM10 for respiratory mortality. A statement that selection used the lowest R² conflicts with conventional interpretation and requires checking in the full text.
02

Why this matters globally

The work illustrates the potential of sequence models for urban-health warning systems if they are validated out of time and paired with epidemiological methods designed for causal effect estimation.

03

Thai researcher contribution

Thai investigators used Bangkok data to connect environmental, engineering and public-health analysis and demonstrated model interpretation with SHAP.

04

Limitations to consider

This ecological prediction study cannot establish causality. Seasonality, influenza, COVID-19, population change and policy may confound results; 2020 overlaps the pandemic. Train-test splitting and out-of-time validation need scrutiny, and SHAP explains a model rather than a biological mechanism.

05

Verify the original sources

Journal of Health Science and Medical ResearchRead the original article

DOI: 10.31584/jhsmr.20261391

KEEP EXPLORING

More Thai research to explore