Chaos Analysis and Machine Learning for Forecasting Climate Change in Some Countries of Latin America

This manuscript aims to assess climate change in some Latin American countries via Chaos Analysis. Specifically, the False Nearest Neighbors method, Lyapunov Exponent, and BDS Test were implemented to study the chaos in a dataset consisting of three climate change-related variables: Primary energy consumption, Total greenhouse gas emissions, and Carbon Dioxide ( \(CO_2\) ) emissions. Throughout this study, we found that the process of climate change in the Latin American countries studied is chaotic, and forecasting values for the following years is a challenging task. However, by using the ARIMA and LSTM models and measuring their performance through MAPE and MSE, we observed that the approximations are not excessively large. In the analysis of these series it was shown that the ARIMA model yields better approximations than the LSTM model. It is important to note that when dealing with chaotic processes, making predictions becomes very challenging, and the errors tend to be high.