Interpretable machine learning approaches to prediction of chronic homelessness

摘要

We introduce a machine learning approach to predict chronic homelessness from de-identified client shelter records drawn from a commonly used Canadian homelessness management information system. Using a 30-day time step, a time series dataset for 6521 individuals was generated, consisting of static features for client attributes and dynamic features describing shelter service usage over time. Five candidate models were trained to predict whether a client will be in a state of chronic homelessness 6 months in the future. The training method was fine-tuned to achieve a high F1-score, with a desired balance between recall and precision, in favour of recall. Mean, recall and precision across 10-fold cross validation were above 0.9 and 0.6 respectively for three out of the five candidate models. An interpretability method was applied to explain individual predictions and gain insight into the overall factors contributing to chronic homelessness among the population studied. This study demonstrates that it is possible to achieve state-of-the-art performance and improved stakeholder trust of what are usually "black box" machine learning models using an interpretability algorithm.