Article Preview
Top1. Introduction
Sports training is a very complex process that usually involves many factors, e.g., coaches, cyclists, sports trainers, trainers' assistants and club officials (similar to team sports). In the past, sports training was performed slightly differently than nowadays, especially because of the rise of technological revolution in pervasive computing. The latter emerged due to the innovative way of thinking, which says that the information should be obtained from everywhere, anytime, for any objective. Monitoring and analyzing various parameters during and after the sports training sessions therefore became a very important tool and the necessary step towards progression. However, cyclists must use various professional sport watches and a bunch of sensors that enable them to track and analyze the whole activity, as well as obtain vital parameters that play a big role in the cyclist's performance (Baca, 2017). A good example of this are power-meters, which are embedded into a bike's frame and are devoted to measure power during the cycling. After the training session, officials and cyclists can make deep analysis of the acquired data and determine how successfully the sports training session was performed and what went wrong. It is worth mentioning that not only professional cyclists use modern tech gadgets, but also amateur cyclists. However, the border line between professional and amateur cyclists is becoming thinner every day. Triathlon is an example of a sport where many amateurs behave like professionals.
In recent years, social scientists began to analyze the behavior and features of cyclists using deep analytics. The following three factors have given rise to the idea of a new paradigm in data processing (Henke et al., 2016):
- •
Data volumes have grown exponentially in the last century
- •
More sophisticated algorithms for data processing have been developed
- •
Computational power and storage have improved steadily
The first factor refers to data collection advances. Nowadays, new data sources allow the generation, calculation and storage of big data volume. These data sources produce data throughout internet searches, social media activities, online transactions, as well as sensors connected into the so-called Internet of Things (IoT) (Whitmore, Agarwal, & Da Xu, 2015). The second factor depends on the fact that the increased data volumes have demanded new self-learning algorithms, instead of the standard hard-coded algorithms. This class especially comprises machine learning algorithms (e.g., deep learning) that use an inductive approach to form a representation of the real world based on the seen data (Henke et al., 2016). The third factor, however, presents a basis for the first two. Computational power has been increased by introducing Graphical Processing Units (GPUs), enabling 10-30 times faster image processing in neural networks, than the traditional Central Processing Units (CPUs). The biggest step forward presents so-called cloud computing, where companies reserve as much computational power and storage volume as needed on the internet without buying any physical machine.