Analyzing the Severity level of Asthma (SLA) using Artificial Intelligence

Abstract

This research represents a notable stride in the evolution of machine learning appli cations within the healthcare domain, seeking to address a conspicuous void in the intricate realm of asthma severity prediction and classification. While extant literature has predominantly concentrated on forecasting asthma attacks, a substantial lacuna persists in achieving a nuanced comprehension and prognostication of individualized asthma severity. The research introduces pioneering methodologies that seamlessly fuse machine learning techniques with the nuanced capabilities of natural language process ing (NLP), leveraging textual statements of symptoms provided by asthma sufferers over an extensive temporal span. The augmentation of the predictive model with the integration of respiratory audio data further enriches the depth and scope of the classi fication paradigm, presenting a comprehensive and holistic approach to asthma severity assessment. The meticulously designed methodology traverses through an intricate process of data preprocessing, a cornerstone element encompassing multifaceted procedures such as to kenization, lowercasing, lemmatization, and the judicious removal of punctuation and stop words. The subsequent utilization of Term Frequency-Inverse Document Frequency (TF-IDF) and Latent Semantic Analysis (LSA) unfolds as a pivotal facet, orchestrating the extraction of features that contribute substantively to the efficacy of the predic tive model. The integration of respiratory audio data imparts an additional layer of complexity, prompting a granular examination of abnormal sounds, including but not limited to crackles and wheeze. Results emanating from the rigorous evaluation process provide a nuanced perspective on the model’s proficiency. A discernible strength surfaces in its capacity to accurately discern instances devoid of abnormalities (’none’ class), while concurrently illuminating challenges in distinguishing between specific classes, thereby underscoring the intricacies inherent in predicting asthma severity based on audio data. The dichotomy between su pervised and unsupervised classifiers unfolds with Random Forest emerging triumphant over Stochastic Gradient Descent in predicting severity. Concurrently, K-means cluster ing manifests as a compelling contender, showcasing comparable accuracy and thereby delineating its latent efficiency in the realm of asthma severity prediction. In conclusion, this research constitutes a substantial and seminal contribution, not merely to bridge extant gaps but to redefine and advance the landscape of asthma management through personalized healthcare. The identified challenges, meticulously dissected within this research, serve as beacons guiding the trajectory of future stud ies, poised to refine models, fortify predictive accuracy, and ultimately redefine the paradigms of personalized asthma management.