Stasiun Cuaca Cerdas dengan Fitur Klasifikasi Cuaca Menggunakan Metode Gaussian Naïve Bayes

Abstract

Weather affects environmental conditions and daily human activities. Rapid weather changes can disrupt activities and potentially cause disasters. Therefore, an accurate and real-time weather monitoring system is needed. However, the high price of commercial weather station devices limits environmental data in many areas. This study aims to develop a low-cost ESP32 microcontroller-based weather station without compromising measurement reliability. This system is equipped with a weather classification module based on the Gaussian Naïve Bayes (GNB) algorithm, which processes three main parameters: air temperature, air humidity, and rainfall, to determine four weather classes: sunny, partly cloudy, light rain, and heavy rain. The GNB model calculates the probability of each class based on the normal distribution of training data, producing fast predictions with low computational load, suitable for ESP32 devices. The sensors used show high accuracy: the BME280 sensor has an R² of 0.9599 for temperature and 0.9994 for humidity, the TSL2561 sensor has an R² of 0.9992, and the tipping bucket has an R² of 1. Weather classification accuracy using GNB reaches 100% based on the confusion matrix. This system provides an affordable and reliable weather monitoring solution for areas not yet covered by commercial weather stations.