Evaluating the application of a computer vision model in the customer service chatbot of an electric utility

With the increasingly application of Deep Learning models in computer vision, data quality and systematic metric assessment have assumed a preponderant role in the model performance. In this paper we present the evaluation approach of a model built to provide a computer vision solution, aiming to visually identify the energy meter box in households able to be connected to the electricity grid. It is provided a discussion on the real need for a careful analysis of the metrics adopted to evaluate the assertiveness of the results obtained, highlighting the evaluation framework, the importance of image quality and the main issues that must be followed to meet such a purpose. The approach herein described turned out to be on a good methodological path, presenting prediction results able to add value to the chosen application and which can be continuously improved, how is it intrinsic to the machine learning process.