An efficient and scalable multi-class machine learning weather forecasting model from multi-variables sensor data

Weather forecasting plays a crucial role in supporting agriculture, transportation, disaster preparedness, and public planning. This research presents the development of machine learning (ML) based models to predict weather conditions using data collected from a mobile weather system deployed in Airforce Base, Kaduna Nigeria. The weather system was built with a microcontroller (STM32F401 32-bit microcontroller) interfacing sensor (BME280), LM2576 DC-DC step-down switching regulator, SD card/adapter, Four-row twenty-column liquid crystal display. The sensor was able to detect and store dataset which includes various meteorological parameters such as temperature, atmospheric pressure, humidity, precipitation, altitude, heat index, and wind speed. The goal was to classify weather outcomes into four categories: cloudy, sunny, windy, or rainy. Four machine learning models—Logistic Regression, Naïve Bayes, Random Forest, and AdaBoost—were trained on this dataset to classify and predict weather conditions. Each model was evaluated using standard performance metrics including accuracy, precision, recall, and F1-score. Among the models, AdaBoost outperformed the others, demonstrating higher predictive accuracy of 100% and robustness in capturing complex weather patterns. The findings underscore the effectiveness of ensemble learning methods, particularly AdaBoost, in developing reliable and scalable weather forecasting systems. This study contributes to the growing body of research promoting data-driven approaches in meteorological predictions