Application of interpretable machine learning in revealing spatial and temporal patterns of dengue disease using climatic factors
Dengue has been a major public health burden in the Philippines. Metropolitan Manila is one of the regions with the highest number of dengue cases in 2016 and ranked with the second-highest increase in dengue cases in 2017. Climatic factors, specifically temperature, precipitation, and relative humi...
Saved in:
| Main Author: | |
|---|---|
| Format: | text |
| Language: | English |
| Published: |
Animo Repository
2022
|
| Subjects: | |
| Online Access: | https://animorepository.dlsu.edu.ph/etdm_bio/10 https://animorepository.dlsu.edu.ph/cgi/viewcontent.cgi?article=1011&context=etdm_bio |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Institution: | De La Salle University |
| Language: | English |
| Summary: | Dengue has been a major public health burden in the Philippines. Metropolitan Manila is one of the regions with the highest number of dengue cases in 2016 and ranked with the second-highest increase in dengue cases in 2017. Climatic factors, specifically temperature, precipitation, and relative humidity provided optimal conditions for dengue vector proliferation and increased dengue transmission. With this, climate-based models developed from machine learning algorithms were often used to provide accurate predictions as it can analyze trends from historical dengue data. Currently, the basis for the predictions of machine learning algorithms is unknown, which is why it is termed a “black-box”. Climate-based random forest models were created through two implementations: randomForest package and ranger package. The ranger package slightly outperformed the conventional randomForest package, making it a better alternative in implementing random forest. Furthermore, models yielded more accurate predictions of dengue incidences with a delayed effect on the datasets. For local and global interpretations, most of the best models had relative humidity as the most influential to dengue incidence in Metropolitan Manila at all spatial scales. For temporal analysis, relative humidity and precipitation followed the trend of the 2018 predicted dengue incidences at a regional- and city scale. The use of model-agnostic methods provided in-depth explanations of random forest models per spatial scale. Lastly, local interpretable model-agnostic explanations (LIME) provided unique insights for random forest models and temporal analysis per spatial scale. Keywords: Dengue; Machine Learning; Black-box; Interpretable Machine Learning; model-agnostic; Metropolitan Manila |
|---|