Ontology-Based Data Collection for a Hybrid Outbreak Detection Method Using Social Media.

Area of Science:

• Epidemiology
• Computational Health
• Data Science

Background:

• Rapidly spreading diseases pose a global challenge, highlighted by the COVID-19 pandemic.
• Online social media platforms offer timely public health information dissemination.
• Traditional disease detection methods can be enhanced by real-time data analysis.

Purpose of the Study:

• To develop an effective infectious disease prediction model using social media data.
• To leverage Twitter data and ontology for identifying and curating relevant disease symptom tweets.
• To improve epidemiological insights and outbreak tracking capabilities.

Main Methods:

• A hybrid model integrating XGBoost and Bidirectional Long Short-Term Memory (BiLSTM) architectures was developed.
• XGBoost was used to handle small datasets and identify optimal features from multivariate time series data.
• Ontology was employed to curate relevant tweets from Twitter data related to infectious disease symptoms.

Main Results:

• The hybrid XGBoost-BiLSTM model demonstrated superior predictive performance compared to state-of-the-art and baseline models.
• Extensive experimentation on a dataset of multiple infectious disease outbreaks validated the model's effectiveness.
• The model showed enhanced forecasting accuracy and improved outbreak tracking capabilities.

Conclusions:

• The proposed hybrid model offers a promising approach for real-time infectious disease prediction.
• This framework can assist health authorities in mitigating fatalities and preparing for future outbreaks.
• Utilizing social media data with advanced machine learning enhances public health surveillance.