Knowledge Content Library

Why Can't I Just Push the Button

Steven Harrod, Lektor/Associate Professor, Technical University of Denmark
July 30, 2021 (SDP Invited Talk)

Abstract: The Danish Railway Authority (Banedanmark and DSB) compensates passengers when a train is tardy beyond a threshold. The rail system collects over two million data points each month to forecast costs, with a typical station amassing over 14,000 data points. In contrast, research on train delays typically analyzes datasets comprising fewer than 1,000 points. The difference is meaningful because, contrary to intuition, the analysis of large datasets can produce misleading results. Complicating matters, data collection, even when automated, adds noise to data, noise that can lead analyses astray. In this talk, Dr. Harrod describes how qualitative factors affect systems and processes and how system data becomes noisy. Using simple computations, he shows how data and noise can mislead data science, especially standard statistical tools, to produce inaccurate results, like poor probability estimates and, for the Danish Railway Authority, significant errors in forecasts of financial liability. Dr. Harrod presents methods for overcoming these problems, helping decision-makers to look beyond data and make better modeling choices.

Click on the file below to hear a sample of the presentation.  

Click here for access to the full video.

SDP membership is required for access to this webinar.

Keywords: analysis and modeling anamod, risk and uncertainty riskunc, distributions, statistics

   Email | Print