Navigating predictive landscapes of cloud burst prediction approaches: insights from comparative research

Abstract

Cloud burst forecasting remains an evolving field that grapples with the complexities of atmospheric phenomena and their impact on local environments. Cloud bursts in hilly regions demand robust predictive models to mitigate risks. This study addresses the challenge of imbalanced cloud burst occurrences, emphasizing the need for accurate predictions to minimize damage. It develops and evaluates a machine learning-based forecasting approach that includes several weather factors such as temperature, humidity, wind speed, and atmospheric pressure. The study also tackles the imbalance in cloud burst data. A dual-axis chart visually merges cloud burst occurrences with weather parameters, providing insights into their relationships over time. The model’s overall accuracy is 0.68, with precision and recall for cloud burst events at 0.25 and 0.07, respectively, and an F1-score of 0.11. However, when it comes to forecasting non-cloud burst occurrences, it shows a high precision of 0.72. This study evaluates machine learning models for cloud burst prediction, highlighting random forest as the top performer with an accuracy of 85.43%, effectively balancing true positives and true negatives while minimizing misclassifications. This research contributes to cloud burst prediction, offering performance insights and suggesting avenues for future exploration.