A new machine learning method for rainfall classification: temporal random tree

Traditional classification algorithms usually assume that all samples in a dataset contribute equally to the training of a machine learning model, which is not always the case. In fact, samples in temporal data, such as precipitation data, may not have equal importance; more recent samples contain more accurate and useful information than earlier ones. To address this issue, the article proposes a novel method, named temporal random tree (TRT), in which recent training samples have a greater impact on the model's decision-making process. It divides the dataset into temporal segments, assigns higher weights to classifiers trained on more recent data, and employs a weighted majority voting strategy. The experiments demonstrated the effectiveness of TRT on the real-world WeatherAUS precipitation dataset, achieving an accuracy of 83.54%, which represents a 5% improvement over the traditional random tree method. Additionally, our method achieved an average improvement of 9.98% compared to state-of-the-art results in the recent literature. These findings highlight TRT's potential as a valuable method for spatiotemporal rainfall classification.