BiDaML in Practice: Collaborative Modeling of Big Data Analytics Application Requirements

Abstract

Using data analytics to improve industrial planning and operations has become increasingly popular and data scientists are more and more in demand. However, complex data analytics-based software development is challenging. It involves many new roles lacking in traditional software engineering teams – e.g. data scientists and data engineers; use of sophisticated machine learning (ML) approaches replacing many programming tasks; uncertainty inherent in the models; as well as interfacing with models to fulfill software functionalities. These challenges make communication and collaboration within the team and with external stakeholders challenging. In this paper, we describe our experiences in applying our BiDaML (Big Data Analytics Modeling Languages) approach to several large-scale industrial projects. We used our BiDaML modeling toolset that brings all stakeholders around one tool to specify, model and document their big data applications. We report our experience in using and evaluating this tool on three real-world, large-scale applications with teams from: realas.com – a property price prediction website for home buyers; VicRoads – a project seeking to build a digital twin (simulated model) of Victoria’s transport network updated in real-time by a stream of sensor data from inductive loop detectors at traffic intersections; and the Alfred Hospital – Intracranial hemorrhage (ICH) prediction through Computed Tomography (CT) Scans. These show that our approach successfully supports complex data analytics software development in industrial settings.