Approach to Predicting Extreme Events in Time Series of Chaotic Dynamical Systems Using Machine Learning Techniques

This work proposes an innovative approach using machine learning to predict extreme events in time series of chaotic dynamical systems. The research focuses on the time series of the Hénon map, a two-dimensional model known for its chaotic behavior. The method consists of identifying time windows that anticipate extreme events, using convolutional neural networks to classify the system states. By reconstructing attractors and classifying (normal and transitional) regimes, the model shows high accuracy in predicting normal regimes, although forecasting transitional regimes remains challenging, particularly for longer intervals and rarer events. The method presents a result above 80% of success for predicting the transition regime up to three steps before the occurrence of the extreme event. Despite limitations posed by the chaotic nature of the system, the approach opens avenues for further exploration of alternative neural network architectures and broader datasets to enhance forecasting capabilities.