Modeling of Dengue-Related Dataset using Artificial Intelligence and Natural Language Processing

In the Philippines, there is a lack of dataset in Cebuano, despite being widely used in the country. This poses a challenge for monitoring global health issues, especially something as urgent as dengue fever. Hence, this study developed a dengue-related dataset in Cebuano. For data gathering, this study used WebHarvy to scrape dengue-related posts from social media platform and conducted an online survey using Microsoft Forms, written in Cebuano. A total of 991 datasets were used for data preprocessing. This study implemented natural language processing (NLP), including tokenization, stemming, stop word and curse word removal, and lemmatization, in collaboration with Cebuano speakers and medical practitioners. Named Entity Recognition (NER) was utilized as the primary model of the study, and the BERT model was employed to assess the dataset’s consistency for training a machine learning model. The results indicate that Cohen’s Kappa Score of the Cebuano annotators were 0.861, implying a high level of agreement. The BERT-LSTM model, on the other hand, achieved an accuracy of 72%. This research aims to contribute a dengue-related dataset in Cebuano to the existing studies of non-English languages and to support the development of future machine learning models for the early detection of potential dengue cases.