Forest Fire Prediction using Liquid Neural Networks

Wildfires pose a significant risk to ecosystems, animals, and people. Predicting them early can help prevent major disasters. In this study, a type of neural network called a Liquid Neural Network (LNN) was employed to predict forest fires based on meteorological and climatic data. The LNN model was tested using the Mendeley Forest Fire Dataset, which includes real data such as temperature, humidity, wind speed, and rainfall. With 96% accuracy, the LNN model predicted the probability of a forest fire. These results show that LNNs can be a powerful tool for early forest fire prediction and could help authorities take timely action to reduce damage. Particularly in areas with dry weather and thick vegetation, forest fires represent a significant threat to natural ecosystems, as well as to property and life on Earth. Early detection and prediction of such fires are crucial for effective disaster management and mitigation efforts. This research explores the use of LNNs, a dynamic and flexible type of neural model, to predict forest fires based on meteorological and climatic data. The model was trained and tested on the Mendeley Forest Fire Dataset, which includes various features such as heat, relative humidity, breeze velocity, and precipitation–factors known to influence fire occurrences. Unlike traditional models, LNNs can adapt more efficiently to changing input patterns, making them well-suited for handling time-sensitive and non-linear data. LNN achieved a prediction accuracy of 96%, resulting in good efficacy in identifying conditions that may lead to forest fires. This level of performance demonstrates the potential of LNNs to support realtime decision-making tools for forest management and emergency response planning.