Rainfall Prediction System Using Machine Learning Fusion for Smart Cities.

Area of Science:

• Meteorology and Atmospheric Science
• Computer Science and Artificial Intelligence
• Environmental Science

Background:

• Accurate rainfall prediction is crucial for daily activities but is increasingly challenging due to climate variations.
• Machine learning offers a promising approach to identify patterns in historical weather data for improved forecasting.
• Selecting the optimal classification technique for rainfall prediction remains a complex task.

Purpose of the Study:

• To propose a novel real-time rainfall prediction system for smart cities.
• To develop a machine learning fusion technique integrating multiple supervised learning models.
• To enhance the accuracy of rainfall prediction in the context of extreme climate variations.

Main Methods:

• Utilized four supervised machine learning techniques: decision tree, Naïve Bayes, K-nearest neighbors, and support vector machines.
• Incorporated fuzzy logic to fuse the predictive accuracies of the individual machine learning models.
• Employed 12 years of historical weather data (2005-2017) for Lahore, including data cleaning and normalization.

Main Results:

• The proposed machine learning fusion framework demonstrated superior performance compared to individual models.
• Fuzzy logic integration effectively combined the strengths of different machine learning algorithms.
• The system achieved high accuracy in real-time rainfall prediction for the studied urban area.

Conclusions:

• The developed machine learning fusion system provides a robust and accurate solution for real-time rainfall prediction in smart cities.
• This approach effectively addresses the challenges posed by extreme climate variations in weather forecasting.
• The findings highlight the potential of integrated machine learning techniques for advanced environmental monitoring and prediction.