In this paper, we summarize our recent research on applying data analytics to a new application area: co-operative education. Many post-secondary institutions currently offer co-operative programs in which students alternate between on-campus classes and off-campus work terms. We observe that the co-operative process produces a variety of interesting data including job advertisements and performance evaluations. We discuss novel data science methodologies we applied to these datasets and the business insights we obtained.

In this paper, we take a new look at the problem of analyzing course evaluations. We examine ten years of undergraduate course evaluations from a large Engineering faculty. To the best of our knowledge, our data set is an order of magnitude larger than those used by previous work on this topic, at over 250,000 student evaluations of over 5,000 courses taught by over 2,000 distinct instructors. We build linear regression models to study the factors affecting course and instructor appraisals, and we perform a novel information-theoretic study to determine when some classmates rate a course and/or its instructor highly but others poorly. In addition to confirming the results of previous regression studies, we report a number of new observations that can help improve teaching and course quality.

We propose a graph mining methodology to analyze the relationships among academic programs from the point of view of cooperative education. The input consists of student job interview pairs, with each student labelled with his or her academic program. From this input, we build a weighted directed graph, which we refer to as a program graph, in which vertices correspond to academic programs and edge weights denote the percentage of jobs that interviewed at least one student from both programs. We show that various properties of this graph have natural interpretations in terms of the relationships among academic programs and competition for co-op jobs. We also present a case study that illustrates the utility of the proposed methodology.

Peak reduction is an important problem in the context of the electricity grid and has led to conservation programs in various jurisdictions. For example, in Ontario, Canada, residential customers are charged higher prices during peak times, while large industrial and commercial customers pay heavy surcharges that depend on their load during Ontario's five peak-demand days. Reducing these surcharges is a challenging problem for large consumers due to the difficulty of predicting peak days in advance.We study the impact of this peak reduction program, called 5 Coincident Peaks (5CP), on consumers by analyzing the difficulty of predicting peak-demand days and peak hours on those days. We find that even the state-of-the art peak-prediction algorithms require consumers to curtail load ten or more times, and even then, they may not identify all five peak-demand days. We also analyze alternative policies that cold help reduce peak demand in Ontario. (C) 2016 Elsevier Ltd. All rights reserved.

This paper reports on the analysis of three years research of undergraduate cooperative work term postings and employer and employee evaluations. The objective of the analysis was to determine the factors affecting student and employer success and satisfaction with the work-integrated learning experience. It was found that students performed better and found co-op placements with an increasing emphasis on leadership in their senior years; however, students rated their first employer the highest. Furthermore, senior students were more successful than junior students in work placements abroad, and extended work terms at the same employer did not increase student satisfaction.

In this paper, we propose a heuristic algorithm for day-ahead prediction of the top $K$ days having the highest peak hourly demand for electricity over a given year. This problem, which arises in the context of critical peak pricing in Ontario, Canada, is difficult because we may have to wait till the end of the year to find out which $K$ days ended up being the peak days. Our solution is to leverage short-term load forecasts and call tomorrow a peak day if it has sufficiently high probability of being a peak day in the time window covered by the forecast. Using Ontario demand data from 2007 till 2013, we show that our algorithm may need to call about 2K peak days to ensure that most if not all of the actual $K$ peak days are included.