Work

MevaL: A Visual Machine Learning Model Evaluation Tool for Financial Crime Detection

Year: 2021
Type: dissertation
Abstract: Data Science and Machine Learning are two valuable allies to fight financial crime,the domain where Feedzai seeks to leverage its value proposition in support of its mission:to make banking and commerce safe. Data is at the core of both fields and this domain, sostructuring instances for visual consumption provides an effective way of understandingthe data and communicating insights.The development of a solution for each project and use case requires a careful andeffective Machine Learning Model Evaluation stage, as it is the major source of feedbackbefore deployment. The tooling for this stage available at Feedzai can be improved,accelerated, visually supported, and diversified to enable data scientists to boost theirdaily work and the quality of the models.In this work, I propose to collect and compile internal and external input, in terms ofworkflow and Model Evaluation, in a proposal hierarchically segmented by well-definedobjectives and tasks, to instantiate the proposal in a Python package, and to iteratively val-idate the package with Feedzai’s data scientists. Therefore, the first contribution is MevaL,a Python package for Model Evaluation with visual support, integrated into Feedzai’s DataScience environment by design. In fact, MevaL is already being leveraged as a visualization package on two internal reporting projects that are serving some of Feedzai’s majorclients.In addition to MevaL, the second contribution of this work is the Model EvaluationTopology developed to ensure clear communication and design of features.
Language: English
Open Access status: green
OpenAlex ID: W3208233167
Cites: 104
Cited by:
Related to: 10