Prediksi Turun Hujan dengan Satuan Per-4 Jam menggunakan Model Convolutional Neural Network (CNN)

Dimas, Imam Hendiasid (2021) Prediksi Turun Hujan dengan Satuan Per-4 Jam menggunakan Model Convolutional Neural Network (CNN). Undergraduate Thesis thesis, Institut Teknologi Telkom Purwokerto.

	Text 1. COVER.pdf Download (278kB)
	Text 3. ABSTRACT.pdf Download (11kB)
	Text 2. ABSTRAK.pdf Download (11kB)
	Text 4. BAB I.pdf Download (141kB)
	Text 5. BAB II.pdf Download (183kB)
	Text 6. BAB III.pdf Download (211kB)
	Text 7. BAB IV.pdf Restricted to Registered users only Download (305kB) | Request a copy
	Text 8. BAB V.pdf Download (11kB)
	Text 9. DAFTAR PUSTAKA.pdf Download (89kB)
	Text 10. LAMPIRAN.pdf Restricted to Registered users only Download (79kB) | Request a copy

Abstract

Indonesia is located on the equator, which makes Indonesia have 2 seasons, namely the rainy season and the dry season. During the transition from the dry season to the rainy season or vice versa, there is an anomaly to the weather, which makes Indonesia generally experience a transition season. Pancaroba causes weather mismatches, such as what should be the rainy season, but in various areas, it experiences drought accompanied by strong winds. This makes weather changes very erratic at the time of changing seasons. The community experienced difficulties in dealing with erratic weather because it can interfere with daily activities and the economy, such as drying agricultural products, making salted fish, shipping, aviation, and housewives. Because it is important to know the weather that will come, to make preparations so that things that are not wanted happen. This research is intended to predict the rainfall at 4-hour intervals using the Convolution Neural Network (CNN) algorithm. CNN was chosen because it has advantages, namely being able to classify an object that is intended for image or image data. The data used are image data taken 4 hours before and after the rain, as many as 640 cloud images, then the augmenter function is performed into 2000 images, divided into 1600 training and 400 for testing. In the implementation of the research, 10 neural network architectural models were made. The best level of accuracy of the results of training or training on these 10 models is 72.44%, namely the 50 (32.64 (64,128)) and 48th epoch. In the testing process, model 50 (32.64 (64,128)) was chosen because it is the most optimum model in data processing, with an accuracy rate of 59%, 26% precision, and a recall rate of 65%. Model 50 (32.64 (64,128)) has an F1 score with a value of 48.10% which is the F1 score with the greatest value among other models. Keywords: Anomaly, CNN, Weather, Epoch, Season, Testing, Training

Actions (login required)

View Item