Forest fires and wildfires represent a significant threat to ecosystems, human lives, and financial resources. Early detection and accurate prediction of forest fire ignition are crucial for effective firefighting. Machine learning methods are emerging as powerful alternatives to complex statistical and physical models in predicting these events. While most research has concentrated on predicting the burned area or the occurrence of fire, our approach focuses on anticipating the initial ignition of fires by leveraging the sequential nature of temporal data. We propose a solution based on both Long Short-Term Memory (LSTM) and Gated Recurrent Unit (GRU) models to capture this sequentially. Additionally, we employ data augmentation and filter out repeated and low-information data. The LSTM model achieved an impressive accuracy of 97.6%, with both models’ performances rigorously evaluated using various classification metrics.