Bronchial asthma, a typical psychosomatic disease, is frequently comorbid with depression and anxiety, while traditional assessment methods have significant limitations in this regard. Machine learning can integrate multimodal data including clinical, behavioral and environmental information to construct accurate predictive models, enabling early identification and intervention. This paper systematically reviews the epidemiological association between bronchial asthma and depression and anxiety as well as their neuro-immune-endocrine comorbidity mechanism, discusses the application potential of traditional machine learning models and deep learning models, emphasizes the core value of multimodal feature integration and model interpretability, and points out the key challenges in current research. This article aims to reveal the value of technological breakthroughs brought by machine learning, and proposes future development directions including the establishment of a multicenter standardized database and the enhancement of model interpretability, so as to provide scientific basis for constructing an integrated psychosomatic diagnosis and treatment system for bronchial asthma and improving long-term prognosis for patients.