Asthma classification using natural language processing and machine learning

Asthma is a non-communicable disease of high importance due to its widespread global presence and impact; it can be described most simply as a narrowing of the respiratory passages, yet it has significant public health consequences for both adults and children, including high morbidity and mortality rates in the most severe cases. Recently, the growing need to expand the use of Artificial Intelligence (AI) algorithms to support automatic clinical diagnosis systems has become more obvious. In this research, therefore, an unorganised database of asthma cases from the University of Grenada was used to extrapolate patients’ asthma conditions. An Iraqi dataset was then collected from an Iraqi hospital with patients suffering from asthma to various degrees, and used to further develop a proposed model in four major stages. The first was data collection and preparation for the mining process. The second was data pre-processing, performed by applying several different natural language processing (NLP) algorithms. The third stage involved features extraction and weighting based on applying a Weighted Term Frequency Inverse Document Frequency (WTF-IDF) approach. The extracted features were then fed into various machine learning techniques to develop diagnoses as the final stage. The classification methods applied were multilayer perceptron (MLP), logistic regression (LR), and linear support vector classifier (LSVC); these were compared to determine which was the most accurate for use in classifying asthma patient data. The findings suggested that the highest accuracy for both the Grenada dataset and the Iraqi dataset (99.89% and 97.51%, respectively) was achieved by applying MLP.