Regresar

Model with NLP and Machine Learning for Predicting Symptoms Based on Textual Conversations with COVID-19 Infected Individuals

Abstract:

AI is a science that has influenced humanity in recent years, with natural language processing being a field within it used in the development of systems that understand unstructured texts and extract essential information from them. This project aims to design a model using natural language processing techniques and supervised machine learning algorithms for the classification and prediction of text based on textual conversations with individuals infected with Covid-19. In the methodology, data was collected through a survey of a population residing in zone 8 of the Guayas province, corresponding to the cantons of Guayaquil, Durán, and Samborondón in Ecuador, resulting in 5000 surveys. Prior to model training, a manual cleaning process was performed, removing irrelevant and semantically meaningless information, including the correction of texts with spelling errors, emojis, and information unrelated to the topic. This process resulted in 4140 records. Subsequently, preprocessing and processing techniques were applied, resulting in 3925 valid records. The modeling utilized a multi-output classifier with the best logistic regression classifier for the classification of multiple labels with binary classes. The results obtained showed metrics including an accuracy of 0.845860, precision of 0.975319, sensitivity of 0.934876, and an F1 Score of 0.953337.