<![CDATA[Airports in Indonesia often face bad weather problems that affect visibility and impact flight operations. Historical data shows several incidents caused by decreased visibility due to fog or rain that resulted in flight delays and cancellations. It can be said that the importance of more accurate visibility estimates to improve safety and operational efficiency at airports. The purpose of this study was to determine the performance of the Back Propagation and Convolutional Neural Network (CNN) models in estimating meteorological visibility at airports because accurate visibility is very important in determining operational decisions, especially during bad weather conditions. The selection of the Back Propagation method is based on its advantages in handling various types of data dynamically and in a directed manner so that it is more precise in predicting visibility based on interrelated meteorological variables. While Convolutional Neural Network (CNN) is very effective in handling problems involving image data. However, currently there are quite a lot of studies that use CNN for text processing because the results are quite promising. The data used is meteorological data that includes temperature, humidity, air pressure, wind speed and other parameters at Radin Inten II Airport. From the results of this study, the Backpropagation model is better in ROC AUC (85%) compared to CNN (84%), this shows a slight advantage in distinguishing classes. The CNN model is better in Precision by 71% compared to Back Propagation 70%, which means it is slightly better at avoiding false positive predictions. CNN has a higher correlation on the test data (0.20) compared to Back Propagation (0.18) indicating its predictions are slightly more in line with the actual data. The larger correlation difference in CNN (0.18) compared to Back Propagation (0.10) indicates a higher possibility of CNN overfitting compared to BP. Since both models show almost the same performance and the difference is not too significant, the choice of model can depend on the specific needs in the implementation. If the goal is to get a more stable model, then Backpropagation is more recommended because it has a smaller correlation difference and higher ROC AUC. However, if what is sought is a model with more accurate predictions in real scenarios, then CNN can be a better choice because it has higher Precision and better test correlation.]]>
Copyrights © 2025
