This study discusses the application value of behavior analysis based on deep learning in the evaluation of depression in art students. Because of the professional characteristics and creative pressure, art college students are at high risk of mental health, among which the incidence of depression is increasing year by year, which has a serious impact on their studies and quality of life. With the rapid development of AI technology, deep learning algorithms show significant advantages in processing complex data and pattern recognition. In this study, by collecting the daily behavior data of art college students and combining it with a deep learning algorithm, an efficient depression evaluation model was constructed. The model aims to realize the early identification and evaluation of depressive symptoms of art college students and provide new methods and means for mental health management. The study collected data using various methods such as questionnaire surveys, mobile application tracking, and social media data crawling, and went through detailed data preprocessing steps, including missing value processing, outlier detection, data standardization, and feature selection, to ensure data quality and model training effectiveness. Subsequently, this study designed a deep learning model (CNN-LSTM) based on the combination of Convolutional Neural Network (CNN) and Long Short Term Memory Network (LSTM), which can capture temporal dependencies and spatial relationships between features in the data, thereby improving the accuracy of depression assessment. The empirical findings demonstrate that the CNN-LSTM integrated model has attained remarkable accuracy in assessing the depressive tendencies of art students, underscoring the efficacy of deep learning techniques in behavioral analysis. This research further scrutinizes the impact of various attributes on the predictive outcomes, highlighting the significance of social interaction frequency, academic stress, and artistic engagement levels in depression assessment.