Individual skills do not change a lot, but with training practice you can reach a level that is hard to classify.
HAGAKURE (XI, 145)
The ancient martial art of Samurais required daily practice and constant discipline, a devotion to warrior arts touching religious fanaticism. “By imitating a good model and practicing with commitment”, said Japanese master Ittei, “even a bad samurai can improve”.
In the Western technological society, industrious and lazy at the same time, the zen spirit of Samurais survives only in the hearts of our modern heroes: the sportsmen. Personalities like Usain Bolt, Lionel Messi or Chris Froome seem to be a perfect mix of talent and determination, technique and intelligence. Their olympic deeds give them the status of champions, efficient machines who convert pain into glory.
Sports and Zen philosophy were precisely the topics that Paolo and I were discussing on a beautiful sunny day, sitting in the comfortable armchairs of the Computer Science department, at the University of Pisa. In particular, we were interested in cycling, the sports Paolo practices almost every day. “There should be a way to understand how to train”, said Paolo, “I think talent alone is not enough”. “Of course it is not”, I replied, “by practicing with commitment even a bad samurai can improve” using the words of master Ittei. The discussion rapidly developed and moved outside, in the middle of the stream of bikes crossing the gracious historical center of Pisa. Many questions began to wander curious in our minds: “Who are the cycling champions? What are the factors determining the success in sports? Is it more a matter of talent or practice?” To be or to do, that was the question.
Just a few years ago, the path of those curious questions would end in the blind alley of mere speculation. Paolo and I do not have access to instruments for measuring sports performance, nor specialized laboratories to observe cyclists. Fortunately, nowadays this is not a problem. To answer those fascinating questions we do not need any expensive laboratory. What we need is for free and available in large quantities. All we need is data.
In the era of social networks, people share on the Web more and more information: emotions, thoughts, photos, achievements, preferences and even their sports performance. The website Strava.com, for example, is a fitness social networking service, a sort of Facebook by which thousands of amateur cyclists like Paolo share their deeds from all over the world, publishing information about the racetrack and the heart rate. Through the services made available by Strava, Paolo downloaded a large amount of data about training sessions of 30 thousands (anonymized) cyclists performed in a period of 7 months, from November 2012 to May 2013. Such amount of data allowed us not only to satisfy our curiousity, but also to perform the first large scale scientific study on cyclists’ performance.
Starting from the Strava data, we tracked the “training history” of each amateur cyclist by computing the effort and the performance achieved in each training session. Exploiting Data Mining algorithms, we divided the cyclists into different clusters, according to the similarity in the performance evolution. Two cyclists belong to the same group if their performance history is similar. Our algorithms unveiled three different clusters: 1) cyclists achieving low performances during all the period; 2) cyclists who strongly improve the performance in the last part of the season (from March to May); 3) cyclists who maintain high performances during all the period. The Figure below gives a graphical representation of the typical performance evolution for the three obtained clusters.
Cyclists in cluster 2 (red solid curve in the Figure), start with the lowest performances. However, they finally reach the best ones in the last and crucial part of the season (May), when the major cycling amateur competitions take place. In those races, they will be the winners. “I definitely belong to cluster 2!”, said Paolo comparing his personal performances with those of our Strava cyclists. “I thought you were with the bad ones”, I replied jokingly, “but there is another aspect we need to understand: How do winner cyclists train?
To answer the crucial question about winning training patterns, we studied the effort history of cyclists in clusters 2 and 3. Surprisingly, we discovered that the effort spent by winner cyclists is generally lower. As we see in the Figure below, a typical cyclist in cluster 2 (the red curve) starts with light workloads, increasing them gradually as the season of the races nears. It seems to follow the well known overcompensation theory: alternating higher loads to stress the body, and periods of rest, during which the body adapts and improves.
Such results leave no room for doubt: not only the engine but also the strategy matters. Cyclists who train in a wise way, alternating stress and rest in the right way, finally reach the best performances. Cyclists who do not, in contrast, lose because their body does not have time to adapt and improve. “You need head and calves”, as legendary cyclist Alfredo Binda used to say in the twenties. Or “… with the training practice you can reach a level that is hard to classify”, as the wisdom of Samurais suggested centuries ago. The data shared by the Strava users allowed a first large scale confirmation of the overcompensation theory. Morever, our analysis revealed that strategy is fundamental even if you are not Eddie Merckx, but just an amateur cyclist.
“Are you going to train today?”, I asked Paolo after the submission of the scientific paper about cycling performance. “Oh no, today I have to overcompensate”, he answered. It was a sunny day in Pisa as the day our adventure in cycling performance began. We walked relaxed around the city, enjoying the short satisfaction that you get after writing a paper. Suddenly, the screams of some kids attracted our attention. They were playing soccer in the street, celebrating the winning goal. Paolo and I silently looked at each other, with a smile of understanding.
The patterns of success in Soccer: challenge accepted.
Luca Pappalardo and Paolo Cintia
You can find details about our research on cycling performance in this paper, published in the “Data Mining Case Studies” workshop of the International Conference of Data Mining (ICDM) 2013 .
Another post about our work was published in the “cycling science” blog here.