Depression has become a common mental disease in society, causing varying menticide and physical damage to individuals and society. The research on depression includes drug treatment, analysis of etiology, forecasting depression and prevention. This paper uses machine learning and deep forest is the core algorithm. In the cascade structure, the method of generating class vectors is redefined, and the forest is reweighted to improve the classification accuracy, which is applied to assist the diagnosis and classification of depression. Age, education level, income, assets and investments are the features to classify whether the individual is depression, the Investigated groups are divided into depression and non-depression. Finally, the classification performance of the improved deep forest is compared with the standard deep forest, random forest, xgboost and logistic. Experimental results show that the improved deep forest has better classification performance in the experiment of assistant depression diagnosis.