 I'm going to introduce you to Kira Feely. You'll already have seen Kira's research being mentioned by Barry. Kira is a final year PhD student in the SFI Centre for Research, Training and Machine Learning based here in UCD. Her research area is around recommender systems applications and her PhD title is using machine learning techniques to support endurance exercise, particularly marathon running. So far Kira has worked on projects and publications around AI models for marathon time prediction, understanding training disruptions and recommending personalised training plans using a dataset from over a half million Strava runners and I think we're going to focus on the training breaks here today so thank you very much. Thanks. Hopefully everyone can hear me but if you can just wave at me so as Brian said I'll be looking at the training disruptions that Barry talked about earlier. So the type of data that we're looking at we've heard about already a bit today so it's all of the sessions that were inputted into Strava between the years 2014-2017 and that is a lot of runners. I'm sure some of you today might have actually contributed to this dataset so you'll see some of your data maybe coming together in some way. So this is the type of data that we have it sort of looks like this every 100 metres we have some information and we also have for some runners we have the heart rate data but not for all runners because around 2014 not everyone was tracking their sessions with a heart rate monitor so we've extracted around one million marathons and the associated training for each of those. This is part of my PhD research which aims to take this type of data and take it from the raw form and try and find a way to represent it so that we can make beneficial supports for marathon runners and at the end of the day what we're hoping to do is make some sort of training plan recommendations that takes into account performance goals and potentially trying to ward off any injuries and everything like that. As Alison mentioned in the sports science literature there's a lot of problems in terms of what's been studied so far in that we often have to rely on these small cohorts that lack elite runners they're either taking place in controlled lab environments or they're relying on runners recall and the Strava data set that we have overcomes a lot of these problems because we now have hundreds of thousands of runners whereas in the previous studies we saw over 85 studies there was a total of 8,500 people so it's quite a difference in numbers and we now have a mixture of elite and recreational runners and it's not at all a controlled environment it's completely just whatever the runners decide to do that's what we see and we don't have to rely on the recall but it comes with its own problems and it's quite an unbalanced data set so we've got a very big difference between the number of males and female runners that we have it's quite noisy so basically I've spent a lot of times looking at these sessions and sometimes I just don't know what a runner is trying to do I've looked at them looking at what their heart rate is doing looking what they're putting a GPS is doing and I can't figure out what their goal was so it's quite a noisy data set and there's also GPS issues because the trackers from around that time weren't as good as they are now so you might be running in Dublin running in Dublin all of a sudden you're in Kansas and now your time has gone backwards and so there's lots of GPS problems as well but the biggest problem with what I'm going to show you today is that our data is unlabeled so we have absolutely no information from the runners about what they were trying to do if the time that they got was what they were going for or if it's very different than what they expected and with the training disruptions that we're going to look at we've no idea why these runners took these training breaks so we don't know is that because they got an injury were they on holidays did they just take their watch off we don't know I'll throw out some baseless speculation later on to go with my analysis but in all we don't actually have any sense of why runners might have taken a disruption so we'll be looking at training disruptions in marathon and in their training programs how does that actually affect their performance consistency is something that we often see as being really really important for marathons for marathon training it's really important to be consistent we'll see that in a lot of online sites but in the studies there's actually no real evidence to show that consistency is important not because it isn't but because it just hasn't really been studied so they'll look at a marathon runners program in terms of how much training they've done so what's their average training what's their total training maybe but not how consistent they were on a week-to-week basis and not how taking larger breaks and trainings might affect their performance that is what we're trying to look at today we're looking at training disruptions what is the frequency of these and what do they cost a runner and here a training disruption is going to be deaf-defined as a period of n consecutive days with a complete success of training so no training took place in this amount of time we did a few things to narrow down our data set so as they said we had a million training programs and we required that there was no GPS issues in their marathon and that their marathon time was somewhere between two and seven hours because it's likely that anything less is maybe a mislabeled session when it was actually a cycling session and anything more it could be somebody who's maybe walking and not running and we also required they had sufficient training and other characteristics to to include and we also required that the program that we were looking at was from their best race in a six month season which stops there being any overlapping training programs especially some runners might be doing marathons as a part of their training but not necessarily doing it as a race so we ended up with around 500 000 training plans from 300 000 unique runners this is sort of how the data is split depending on males and females and in terms of the race times so we have around four times as many males as female runners in the data set which is sort of similar to what we find in the studies that we've seen so far and they're mostly between three and five hour marathon times so we're first looking at what is the the frequency of training disruptions so how often do people take a long training break in their marathon training programs so we can see that on the left here at the moment this thing will work yeah okay around 60 percent maybe a little bit less 55 percent of runners have a training disruption of at least seven days at some point in the time three to 12 weeks from race day so they've taken a full week off consecutively of training in that period three to 12 weeks from race day and still managed to go ahead and complete a marathon three to 12 weeks later so I think that's kind of a good sign if anyone's worried about taking a week off in training so many runners do this and still manage to actually complete their race which is a good sign and even more so around uh here we are around 15 or so percent of runners will take a break of at least 14 days and still manage to complete a race three to 12 weeks later which I think is a good sign and there is a difference between males and females it's not much but it is statistically significant up to breaks of at least 21 days and same thing goes for younger versus older runners where we see a more dramatic change is in the sort of looking at whether the break happens early in training or later on in training so people are more likely to have a longer training disruption earlier on in their training so eight to 12 weeks from race day compared to closer to race day at three to twelve three to seven weeks basically but we can see we still have around 10 percent of runners having a 14 or more day disruption in the three to seven weeks before race day and still going ahead and doing a marathon a few weeks later which I think is pretty good we also see that we have more disruptions happening for faster runners versus slower runners which is interesting and I hope I'm not offending anyone when I say more than four hours is a slower runner because if I tried to run a marathon it would take me a lot more than four hours but yeah that's just how we're sort of defining it in this work we also want to look at how does a disruption of a certain length affect your marathon performance so I looked at what was the max disruption length that someone had in their three to twelve weeks before race day and what was the average race time of these groups and you can see for people whose max break is very low around two or three days their average race time is 205 minutes so this these are people whose max disruption in the whole of their training is two consecutive days off which is nothing compared to people who are even just around seven days their average time jumps up to around 250 minutes which is quite a big difference so as we might have heard before correlation does not imply validation so looking at this doesn't mean necessarily that having a disruption of a longer time means that your performance is going to be impacted it could actually be the other way around that people who are slower tend to take more disruptions or there could be something we have no idea going on like people who are slower have more holidays and tend to just take more time off we don't really know what could be going on there but to investigate this further we decided to look on a per runner basis what might be the performance costs associated with a training disruption so we took all of the runners who had run at least two programs one of which was undisrupted so their maximum disruption was less than seven days and one of which was disrupted so they had at least a seven-day break at some point in their training program and we compared the finished times of those two programs on a per-runner basis to see for an individual what is the actual performance cost that's associated with having a disrupted training program in terms of how many runners that was so out of the 300,000 programs or the 300,000 runners and 500,000 programs we're narrowing it down quite a lot to find these runners so we're looking at about 40,000 for disruptions of seven to 13 days and even less kind of 10,000-5,000 for longer disruptions so we do lose out on a lot of runners by doing this per-runner analysis what we're looking at here is the percentage difference in marathon times when we're looking at a disrupted program versus an undisrupted program so how much does the disrupted time add to your marathon basically so if we had a value of five percent here it would mean that your disrupted program leads to a five percent slower time than your undisrupted program so on average a disrupted program leads to a four to seven or eight percent disrupt performance cost compared to when you're following an undisrupted program so that's just looking at the average and we can see because we have most runners in the seven to thirteen days the average is really around five percent that's kind of what you can expect on average the performance cost might be if you have a disruption and we can see there is a difference between whether the disruption happens before the undisrupted program or whether the disrupted program happens after and it's hard to know is that just that having knowing that we've had a disruption before tells us a bit about what type of runner that might be maybe there's generally more inconsistent and that might mean that it has a greater cost it's hard to know when we look at the difference between males and females sort of similar to some of what Barry showed us earlier for males again we have a greater performance cost compared to female runners the males are looking at somewhere between four and eight percent and for females it's sort of lower than four percent same thing when we're looking at older versus younger runners we have a greater performance cost for younger runners now this could be because we tend to have more faster runners in the male and the younger category which are contributing to to the overall makeup and we'll see in a second that there is a greater performance cost significantly for faster versus slower runners and I think I think the reason for this might be we don't know for sure and that's something we'll need to investigate further in sports science that's sure but I think the reason that we're seeing the faster runners having a greater performance cost is that when faster runners take some sort of disruption it's likely that it's for a good reason like maybe they do have some sort of injury whereas the slower runners and especially more novice runners or more recreational runners will take disruptions for many different reasons and so maybe they're just generally a bit more inconsistent in their training again we see a greater performance cost when our training disruption happens late in training so closer to race day compared to earlier race day and that might be expected so this is sort of saying okay disruptions are not great most people will have a five to eight percent disruption cost associated with it but the data is very variable so we can see that if we're looking at everyone and not just the averages it's really really mixed how people are going to have disruptions or how much it's going to cost them so on average we're looking at around five percent but there's a number of people who have a disrupted program and have run a significantly faster marathon compared to when it's undisrupted so basically in general they're probably not good to do wouldn't recommend them but a lot of people will not be affected by a disruption a lot of people will run a significantly faster marathon afterwards so in conclusion training breaks are very frequent particularly for slower runners and particularly further back in training many many runners will have some sort of disruption and still go on to complete a marathon so if you're worried about your training that you've had to take a break for whatever reason it's probably not going to cost you the ability to do your marathon but there probably will be some performance costs somewhere around five to seven percent but there is a lot of inter-runner variability and so it won't cost you the race but it might cost you a few minutes that's essentially the takeaways from that so thanks very much