Outlier detection

About the Project

At Appfolio we have been solving multiple problems in the Property Management Accounting domain with Machine Learning technologies. Many customers do accounting with our AppFolio Property Management (APM) solution, which means that they need to input all their records (invoices, bills, checks, etc.) in Appfolio. If the customer has integrated their bank account with Plaid (third-party provider) in APM, we also have data coming directly from their bank transactions.

Customers have been telling us that they would appreciate it if they could detect anomalies in their records and transaction history. For example, they would like to know if the water bill is much more expensive than the previous months for one property, since that would likely indicate a problem. They would also appreciate knowing if there’s a bill that usually comes but it doesn’t. This is a general problem with applications in multiple areas inside our software, and so we would like to have a way of detecting anomalies automatically so that property managers don’t have to worry about every single detail for each property.

For this project, students will analyze outlier detection methods, performance evaluation metrics for empirical comparisons between methods, and assess the methods from an explainability perspective. Students will develop comprehensive method comparisons and integrate them in a dashboard to present them with the correct data visualization techniques.