IMPLEMENTASI ARTIFICIAL INTELLIGENCE DALAM PEMANTAUAN DAN PENGONTROLAN DEMAM BERDARAH
Dengue Haemoragic Fever (DHF) is a disease that is often found in tropical and subtropical climates and will generally increase during the rainy season. Indonesia, which is a tropical country, is also one of the countries with a high number of DHF cases every year, even in 2016 it was the country wi...
Saved in:
| 主要作者: | |
|---|---|
| 格式: | Final Project |
| 語言: | Indonesia |
| 在線閱讀: | https://digilib.itb.ac.id/gdl/view/79644 |
| 標簽: |
添加標簽
沒有標簽, 成為第一個標記此記錄!
|
| 機構: | Institut Teknologi Bandung |
| 語言: | Indonesia |
| 總結: | Dengue Haemoragic Fever (DHF) is a disease that is often found in tropical and subtropical climates and will generally increase during the rainy season. Indonesia, which is a tropical country, is also one of the countries with a high number of DHF cases every year, even in 2016 it was the country with the second highest DHF cases in the world after Brazil. Until the end of 2020, there have been at least 108,303 cases of DHF followed by 661 deaths due to the disease in Indonesia. DHF is caused by the dengue virus which is mediated by mosquitoes. The types of mosquitoes that become dengue vectors come from the aedes genus such as Aedes aegepty and Aedes albopictus.
In this study, a mosquito egg counting system was developed using Deep Learning based Convolutional Neural Network (CNN) in the form of You Only Look Once (YOLO) model and Faster Region Based Convolutional Neural Network (Faster R-CNN). The modeling flow begins with data preparation. Then the hyperparameters are determined on both models, then model testing, and end with model comparison.
After modeling and comparing the two models, it was found that by using the SGD and AdamW optimizers, both models were able to learn the data well. The YOLO model with the SGD optimizer produces an average accuracy of 97.14% while with AdamW produces an average accuracy of 98.85%. The Faster R-CNN model with the SGD optimizer produces an average accuracy of 94.4% while with AdamW produces an average accuracy of 95.6%. The use of AdamW optimizer is considered to provide higher accuracy in both models than SGD. Learning rate variation only makes a difference to the SGD optimizer of the two models. While the batch size variation does not have a big impact on the accuracy of the model. This accuracy value increases compared to the use of the connected components method in previous studies which can only provide the highest accuracy value of 91.3%.
|
|---|