Oilers announce winner of Hackathon 2.0
|Oilers Hackathon 2.0 was launched in December 2012.
View Final Rankings
EDMONTON - The results are in and a winner has been chosen.
Launched last December, the Hackathon 2.0 saw Oilers open up their vault of in-depth and historical data, welcoming contestants to submit their own analyses and present a compelling case to help the club win games.
Measured against real game results, both with the Oilers and across the National Hockey League, entries were scored over the course of the 2012-13 season based on their accuracy in answering four queries posed by the Oilers Analytics Working Group:
- Predict the regular season's points per game for the players listed.
- Predict the season's even strength save percentage for the goaltenders listed.
- Predict the goal differential per regular season game (goals for less goals against divided by games played) for all 30 teams in the 2012-13 season (worth 20 points).
- "Conduct a predictive analysis of your choice on some dimension of potential value to the Oilers. The analysis must be testable in the upcoming season and judged on its difficulty, accuracy, clarity, and value."
"We were extremely excited about all the thoughtful submissions we received from everyone who entered the Hackathon," said Nick Wilson, a member of the Oilers Analytics Working Group and Chair of the Hackathon 2.0.
"Michael's entry was particularly unique. The amount of time and dedication he put into his methodology was outstanding. He went above and beyond. His entry was robust, thought-provoking and highly creative."
Broken down to the bare essentials, Parkatti's submission measures the effect of coaching strategy on standing points, which looks at three main points of study concerning line matching, zone starts and line composition.
"I've always been interested in coaching strategies, so my goal heading in was to see if there was a number that I could come up with to compare various strategies between coaches across the NHL and see if it had any effect winning games," he said.
Parkatti posted the problem by combining three metrics to determine how a coach is composing his lines, where their shift is starting and who they're deployed against.
- Opposition Line Matching - Corsi Relative Quality of Competition (QoC): This measures the shot attempt differential (per 60 minutes) of opposing players relative to their own team that a certain player faces. It's a proxy for ‘how good' the competition a certain player plays against is.
- Zone Starting - Offensive Zone Start %: This shows what percentage of a certain player's shifts started in the offensive zone.
- Internal Line Composition - Corsi Relative Quality of Team (QoT): This measures the shot attempt differential (per 60 minutes of ice time) of the teammates a certain player plays with.
"The idea is to combine all those different ideas to create the Coaching Activity Index (CAI), which, ultimately, is an average of the three other criteria."
Parkatti calculated the Coaching Activity Index for all 30 teams over the past four seasons in order get a sense of which teams employed a line-matching strategy. His hypothesis was that a higher CAI score correlates positively with a team's standings points.
"I was looking for statistical significance and a relationship between the two," he said. "By the end of the (2012-13) season, it was pretty clear that this year, like every other year I calculated, showed fairly high statistical significance between the metric I created and how teams placed in the standings."
"Our criteria in evaluating Question 4 was cross-referencing those conclusions with the results obtained in Questions 1 through 3," said Wilson. "The biggest thing we were looking for was how the idea could potentially help the Oilers' on-ice performance moving forward.
"Michael's entry did particularly well in showing that."
Graded head-to-head over the past four months, Parkatti's entry was the strongest, but other Hackathon entrants submitted some compelling arguments, too. Other proposals included in-depth looks at overall player ratings, why Player X should or should not be drafted to begin with, special teams success (or lack thereof), year-over-year deviations and so on.
The analytical approach is rapidly evolving, meaning the NHL and its clubs are now entering a whole new world of coaching and pro scouting.
"I believe analytics are extremely important in our sport," said Parkatti. "We're starting to get a sense of what kind of things are important and that's a vital step in the process of understanding how teams win hockey games. I think there will be competitive advantage for the franchises that do this first and do it well.
"To me, I think a lot of it is finding value in players. In a salary cap world, you can't simply buy your problems away. You have to find value in your contracts, in your players and find value where others perhaps haven't. A lot of people are potentially fooled by what they're seeing with their eyes and they haven't truly found certain aspects of the game.
"Unlocking those insights is very important."
According to Parkatti, the only stumbling block is the availability of statistical information. What the NHL provides is a good start, but there's more waiting to be uncovered.
"The League publishes very detailed game stats and have for the past several years, but it doesn't tell you everything. It only tells you a defined series of events that happen throughout a game, but I think there's been some good innovation by various people in the hockey analytics community that have been trying to fashion new ideas that aren't currently being published by the NHL.
"The potential is infinite, but it comes down to the collection of data in the end. That's an exercise in how much you want to invest to draw these insights out."
That's why the Oilers have developed an Analytics Working Group -- a staff backed by Dan Haight and Edmonton-based Darkhorse Analytics, who has been working with the Oilers since the 2004-05 work stoppage.
"Our goal with the Hackathon was to access great talent and to come up with great ideas -- with the former being most critical," said Haight, who manages a research centre at the University of Alberta. "This space is very competitive and it's only going to get more and more competitive in the years ahead. Any edge we can get in getting talented people who understand analytics in hockey, we're going to be better off. The best thing we got out of this was access to some bright minds, bright ideas and connecting with that talent moving forward."
As a result of winning the Hackathon, Parkatti will now be working with the Oilers Analytics Working Group, either on a part- or full-time basis.
"There are great hockey minds, people who have been around the game, great scouts, who will be watching a game and will get a gut feel about a player," said Haight. "All that gut feel is, is that they're subconsciously collecting data points. 'That guy has good hockey sense because he made that good play and that good play and that good play.'
"They're saying it's gut feel, but it's the same thing as what we're trying to do. A lot the time it's about watching the play behind the puck or the game within the game. The sport of hockey moves so quickly that you can't see everything happen at once. There's where you need bright minds and intelligent data collection to benefit."
When it comes to the Hackathon, participants were given the avenue to succeed by having access to information that had previously been kept under lock and key.
That, Parkatti says, opened the door even further.
"The data that we received was probably my favourite part of taking part in this process," he laughed. "We got our hands on very complete data sets with a whole variety of subjects that would have taken me weeks or even months of my time to compile. It was awe-inspiring to literally see millions of records of hockey data.
"It expanded the potential of things that I could have a look at. And I think it's expanded the potential of where our sport is headed."
-- Ryan Dittrick, edmontonoilers.com | Follow me on Twitter @ryandittrick