Qualitative activity recognition using machine and deep learning: Experimenting with data-human interfaces for non data-scientists

Data science has become more and more powerful as the development of algorithms and computing power has made huge progresses. Researchers have used machine learning and deep learning algorithms in order to solve hard problems in a variety of domains. Nevertheless, data science will only show its true potential the moment that all categories of interested parties (not only specialist data scientists) will be capable of understanding and interacting with data. This work, hence, is not about data science in a classical way, but about exploring the data-human interface that may be possible to establish utilizing available tools, with little or no background knowledge of how a data science pipeline works. In this paper we show how, within a Sports Science degree course, where no specific knowledge regarding algorithms, statistics or data science is acquired, it was possible to obtain significant results in relation to a data analysis problem, using off-the-shelf application packages. In particular, we here show how, without any specific customization, it has been possible to employ the machine and deep learning algorithms offered by the publicly available platforms, to solve a fitness exercise classification problem, obtaining performances that would have been deemed remarkable until not long ago.