<![CDATA[Daily rainfall is an important climatic element for agriculture, water resource management, and hydrometeorological disaster mitigation. The temporal variability of rainfall makes its classification a challenging task. This study aims to develop a daily rainfall classification model using a one-dimensional Convolutional Neural Network (1D CNN) based on rainfall (RR) data from the Climatology Station of D.I. Yogyakarta for the period from July 1, 2025, to June 30, 2026, comprising 364 observations. The classification uses two classes, namely No Rain (RR = 0 mm) and Rain (RR > 0 mm), with a 30-day sliding window. The 1D CNN architecture consists of two convolutional blocks, batch normalization, ReLU, max pooling, global average pooling, a dense layer, L2 regularization, and dropout. Evaluation on 67 test samples yielded an accuracy of 74.63%, balanced accuracy of 66.84%, macro F1-score of 0.671, and ROC-AUC of 0.777. However, five-fold TimeSeriesSplit resulted in a balanced accuracy of 50.00%. These results indicate that the model performance is not yet stable due to the limited one-year dataset and the use of a single predictor variable.]]>
Copyrights © 2026
