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

Area of Science:

• Machine Learning
• Data Science
• Environmental Science

Background:

• Traditional classification models assume equal sample importance, which is often inaccurate for temporal datasets.
• Recent data in time-series, like precipitation, typically holds more relevant information for current predictions.

Purpose of the Study:

• To introduce a novel Temporal Random Tree (TRT) method that weights recent data more heavily in machine learning models.
• To enhance the accuracy of spatiotemporal rainfall classification using time-aware data weighting.

Main Methods:

• Developed the Temporal Random Tree (TRT) algorithm, which segments data temporally.
• Assigned higher weights to classifiers trained on recent data segments.
• Implemented a weighted majority voting strategy for final predictions.

Main Results:

• TRT achieved 83.54% accuracy on the WeatherAUS precipitation dataset, a 5% improvement over standard random trees.
• The method showed an average improvement of 9.98% over current state-of-the-art techniques.
• Demonstrated superior performance in spatiotemporal rainfall classification tasks.

Conclusions:

• The Temporal Random Tree (TRT) method effectively leverages the importance of recent data for improved classification.
• TRT offers a significant advancement for spatiotemporal rainfall prediction and classification.
• This approach has substantial potential for applications in meteorological data analysis.