Analysis and prediction of PM2.5 in Medellin based on seasonal time series*

Exposure of people to high levels of air pollution can lead to health problems, and consequently a reduced quality of life. By being able to estimate future pollution levels, preventative actions can be taken to avoid exposing people to dangerous levels of air pollution. This article focuses on forecasting the average daily concentration of fine particulate matter (PM2.5) in the Aburra Valley (Colombia) over a one-year period. Leveraging data from the Aburra Valley Early Warning System (SIATA), we explore two predictive models: a Python-based Artificial Neural Network (ANN) and an Autoregressive Integrated Moving Average (ARIMA) model. Our analysis capitalizes on seasonality patterns identified within the SIATA dataset. The parameter estimation of our models is based on the autocorrelation function (ACF), the augmented Dickey-Fuller unit root (ADF) test, and the partial autocorrelation function (PACF). Besides, the performance of our models is evaluated using the Root Mean Square Error (RMSE) and Mean Absolute Error (MAE) metrics.