Prediction of Carbon Monoxide (CO) Atmospheric Pollution Concentrations with Machine Learning and Time Series Analysis in Langkawi, Malaysia

Abstract

Carbon monoxide (CO) is a non-irritant toxic and odourless gas produced from the incomplete combustion of fossil fuels. Long-term exposures to lower levels of carbon monoxide have wide implications for human health. Thus, an early warning system for CO atmospheric concentration with an accurate and reliable forecasting method is crucial. Studies for predicting CO atmospheric concentration are still limited in Malaysia especially using data science approaches. This study aims to develop and predict future CO concentration for the next few hours by using the statistical time series approach and machine learning approach. The data used for the project is the air quality data of the monitoring station in Langkawi, Malaysia. The data mining tool used for this project is RapidMiner Studio. Based on the results, it showed that Time Series analysis with deep learning gave a reasonably good CO concentration prediction for the next 3 hours with a relative error of approximate 10%. The model developed in this project can be used by authorities as public health’s protection measure to provide an early alarm for alerting the Malaysian populations on the air pollution issue.

Keywords

• machine learning
• time series analysis
• carbon monoxide
• prediction