Logistic Regression Predicting the Odds of a Homeless Individual being approved for shelter

Abstract

In this research, we use a logistic regression model to identify the probability of a homeless person being admitted to shelter accommodation depending on the demographic and behavioral factors. The data of 242 homeless applicants acquired through Kaggle was used to analyze nine predictor variables, such as age, gender, veteran status, monthly income, number of nights homeless, substance abuse, completion of in-house training, probation status, and type of assistantship. Following data cleaning and model selection process, five noteworthy predictors were gathered, namely, veteran status, monthly income, number of nights homeless, substance abuse, and type of assistantship. The performance of the model measured by the Hosmer-Lemeshow test, McFadden R 2, and ROC-AUC, showed that the model was well-fitted and predictive (AUC 0.90). The results indicate that homeless veterans, those who receive temporary assistance, those who experience more years of homelessness and their substance abuse problems are more likely to be approved of shelter and higher-income applicants are less likely to be approved. Such lessons can inform the policies of the public health and management of shelters in allocating resources to vulnerable groups in the homeless population.