Toward an Effective Identification of Tweet Related to Meningitis Based on Supervised Machine Learning

Abstract

Epidemic surveillance requires a rapid collection and integration of data and events related to the disease. Adequate measures, including education and awareness, must be rapidly taken to reduce the disastrous consequences of the disease. However, developing countries, especially those in West Africa, face a lack of real-time data collection and analysis system. This situation delays the analysis of risk and decision making. The aim of this research is to contribute to the surveillance of the meningitis epidemic based on Twitter datasets. The approach, we adopted in this research is divided into two parts. The first part consisted of investigating different methods to convert the tweet data into numerical data that will be used in machine-learning algorithms for the classification tasks. The second step is to evaluate these approaches using different algorithms and compare their performance in term of training time, accuracy, F1-score, and recall. As a result, we found that the SVM machine algorithm performed good with 0.98 of accuracy using the TF-IDF embedding approach while the ANN algorithm performed good with accuracy of 0.95 using the skip-gram embedding model.