Differences between descriptive analysis (big data) and predictive analysis (AI)
The ladder of artificial intelligence, as shown in Figure 1, is divided into 4 phases that allow data to be transformed into knowledge. The first two stages (collection and organization) relate to the descriptive dimension of the information where it can be made analysis.
However, with descriptive analysis it is no longer possible to develop phase 3 (analysis) where the question “why did it happen?” is answered, since this requires modelling the relationship between variables, something that is only possible through the use of machine learning.
Finally, phase 4 (knowledge) relates to the predictive (What is going to happen?) and prescriptive (What can I do to achieve the goal?) dimensions, and consists of the application of artificial intelligence models to optimize the processes of an organization (club).
- Maximum uncertainty
- Contextualised future performance
- Minimal uncertainty
- Scientific validation
The artificial intelligence models used in Olocip are scientifically validated using standard procedures in machine learning.
For this purpose, data are separated into training, development and test sets (Figure 4). The candidate models are learned from the training data, the best candidate is selected according to the development data and his ability to generalize on the test data is evaluated.
In the attached image (Figure 5) can be seen the difference between past information (big data) and contextualised predictive information (AI). These are real examples from this 2019/2020 season extracted from our predictive model developed with machine learning with a reduction of the mean square error with respect to the descriptive approach by 48% in the test data.
So, if the question is how would a player perform if we sign him, and only previous data is considered, the obtained information will be deficient. In addition, it is not right to make predictions of player’s performance in future context that are only based on past data and past context. Thus, it’s fundamental to contextualise before comparing and only artificial intelligence is capable of that.
It is important to ensure that the question we ask ourselves is answered and not another. Not everything that is shown as artificial intelligence is. To be sure, it is necessary to ask for references and scientific validations. Predicting performance can and should be done if you want to have a competitive advantage.
More information in email@example.com