Application of machine learning on depression prediction and analysis

Abstract

In this paper, machine learning methods based on Python are applied to predict and analysis the possibility of depression, which provide the possibility whether the individuals suffer depression. Four kinds of machine learning methods, including, supported vector machine, naïve Bayes, random forest and neural network, that make predictions are all based on 23 types of features from the public depression dataset and after processing the data, some relevant figures have also been drawn. The results show that the random forest obtains the best performance among all the methods. Random forest achieves 86.8% accuracy, which validates the effectiveness of machine learning based depression prediction. Furthermore, the important factors which lead to depression are also analysed. Among the predictions of the depression dataset, these figures show that the most influential features are the age, level of education, living expenses and the gender does not have any effect.