Evaluation of neural network feature and function settings on the model performance and accuracy
| Autoři | |
|---|---|
| Rok publikování | 2023 |
| Druh | Článek v odborném periodiku |
| Časopis / Zdroj | Journal of Physical Education and Sport (JPES) |
| Fakulta / Pracoviště MU | |
| Citace | |
| www | https://www.efsupit.ro/images/stories/aprilie2023/Art%20123.pdf |
| Doi | http://dx.doi.org/10.7752/jpes.2023.04123 |
| Klíčová slova | ANN; DNN; inputs; outputs selection and extraction; hidden and output layer activation function; optimization algorithm |
| Popis | Both in general and in sport, the neural network is one of the most frequently used type of the artificial intelligence algorithms. Due to their high price, wealthy sports clubs could only afford to use it on a daily basis in the past. The situation has changed with the development of affordable softwares and their manuals. Although sports club managers can employ them more nowadays, there is still a problem how to prepare the data ideally and to set up the most efficient model algorithm. In the executed studies and literature sources, the function setting has been examined and emphasised more, whereas the set-up of the feature setting has been rather neglected. It has only been recommended to improve the model, but there have not yet been sufficient observation. The current study aims to determine if the features or the function settings have a greater effect on the model accuracy. The initial feature dataset (n = 18882) was obtained from publicly available sources. Each of the six different feature settings consisted of 96 models. A total of 384 models were created, in which their testing accuracy and the percentage difference between the training and the testing phases were further analyzed. No statistically significant differences were found in the accuracy of the function settings, but statistically significant differences were found in the feature settings. Based on the results, the current study concludes that the feature settings are a more important factor to increase the model accuracy than the function settings, especially the reduction of the number of the outputs. Furthermore, the study found that the variables Weight, Height, and Age had the highest frequency of the occurrence of the normalized importance. Therefore, they can be identified as among the most important features to predict the final rank. The results of this study suggest that more emphasis should be put on the feature setting and not just on the function setting when preparing the first model of artificial intelligence. |