Comparison of Genetic Algorithm Optimization with Support Vector Machine (SVM) for Weather Forecast

Abstract

Weather forecasts are one of the important factors for daily activities. It can be used for daily work activities such as farming, aviation, production and distribution. The Meteorology, Climatology and Geophysics Agency makes weather forecasts based on weather parameters, namely temperature, air pressure, solar radiation, humidity and rainfall. The weather forecast class is divided into 5 classes, namely Cloudy, Rainy, Sunny, Cloudy Rainy and Cloudy Sunny. In this research, a comparison of weather forecast models using Learning Vector Quantization optimization and Genetic Algorithms will be made with weather forecast models using the Support Vector Machine method. The data used in this research is weather data at the Citeko Class III Climatology and Geophysics station, the data used is data from the last 3 years. Then the data is divided into training and test data using percentage split with a division of 65% used for training data and 35% used for test data. After making the model using the LVQ-GA and SVM methods, a comparison of the model test results was carried out, from the test results the accuracy value was calculated using a confusion matrix for each model. The accuracy result of the LVQ-GA optimization weather forecast model was 73%, while the weather forecast model using the SVM method obtained an accuracy value of 81.5%, thus the results from SVM were better.