Analyzing & Predicting Depression Risk & Types

Abstract

This research is based on prediction of depression risk, analyzing the initial mental status of a patient and depression types by combining the knowledge of computer science, data analytics with the medical field. Depression is a one of a leading cause of disability worldwide. It’s a mental illness that effects negatively on how you feel the way you think and how you act. In this research, we considered 4 types of depression (Major, Persistent, Bipolar, and Atypical) out of all main 6 types of depression. Here, we selected 1230 people for this research and collected data related to 28 attributes of depression were selected as variables. Statistical techniques were applied to predict the risk of being a patient and to analyze the dataset. Statistical analysis was done for identify the effectiveness of each risk factor on the depression prediction and the most suitable risk factors (p value <0.05) were identified and visualized based on the target variable (patient/not a patient) attribute. Statistical model has created by applying binary logistic regression model. 2 mathematical equations (to calculate Y’ and probability (P)) are consisted in the created statistical model which provides the probability of having depression for any person (depression positive range >0.5, depression negative range< 0.5). Model accuracy is 92%. Hosmer-Lemenshor test shows that the model fts the data well hence the value is 0.486. A system was developed by using the result of the depression prediction with a user friendly UI to find the risk of having depression or not. Initial metal status of a person; high, medium, low was analyzed using both clustering (k-means) and classification (Naïve Bayes) to identify the best suited model for the dataset. Naïve Bayes model accuracy is 70% and the accuracy of the k-means clustering is 55% percent. To predict the depression type, decision tree (J48 tree) was used and tree was built with 31 number of leaves and 48 as the size of the tree. Accuracy of the decision tree analysis is 72% and kappa statistics takes value is 0.6397 which indicates the good performance of the model.