K-NEAREST NEIGHBOR TO PREDICTE WEATHER CONDITIONS USING RAPIDMINER

Abstract

The use of machine learning technology for weather prediction has become an increasingly important approach in modern meteorology. This study implements the K-Nearest Neighbor (KNN) algorithm to predict air temperature using a dataset from the High-Performance Wireless Research and Education Network (HPWREN). The dataset covers a period of three months with various weather parameters such as air pressure, wind direction, wind speed, and relative humidity, while the rain accumulation and rain duration variables are excluded because they do not have significant values. The research methodology consists of four main stages: data collection, preprocessing using RapidMiner, KNN implementation, and model evaluation. The preprocessing stage includes data normalization and selecting relevant attributes to ensure the quality of model input. The KNN model is configured with optimal parameters, namely k = 5, weighted vote, and MixedEuclideanDistance to improve prediction accuracy. The dataset is divided into training data (70%) and testing data (30%) to validate model performance. Model evaluation using three standard metrics yielded a Root Mean Squared Error (RMSE) of 3,680, Mean Absolute Error (MAE) of 2,216, and Root Relative Squared Error (RRSE) of 31.5%. These results indicate that the KNN algorithm effectively predicts air temperature with a level of accuracy sufficient for practical applications. This study provides a foundation for the development of more accurate weather prediction systems in the future, with recommendations to expand the data collection period and further optimize model parameters.