To train the system, the researchers used data from coal mines in China and compared 10 machine learning algorithms to determine which AI method would be best capable of making predictions about changes in methane gas levels a half-an-hour in advance, and subsequently notify users of any anomalies.

The researchers determined that of the 10 machine learning algorithms examined, four reportedly produced the best results.

"Linear Regression is one of the most efficient algorithms with better performance for short-term forecasting than others. Random Forest frequently shows a statistically lower error performance and achieves the highest prediction accuracy. Support Vector Machine performs well and has a shorter computational time on small datasets but will require too much training time as the dataset size increases. The findings of this study will help the coal mining industry to reduce the risk of accidents such as gas explosions, safeguard workers, and enhance the ability to prevent and mitigate disasters which will lead to financial losses in addition to potential losses of lives," the researchers explained.

The team added that the technology could also be fine-tuned for other industries including aerospace, oil and gas and agriculture, among others.

The study, "Comparative study of ten machine learning algorithms for short-term forecasting in gas warning systems" appears in the journal Scientific Reports.