Performance analysis of various features and functions settings on the neural network accuracy

BOZDĚCH, Michal. Performance analysis of various features and functions settings on the neural network accuracy. In -13th INTERNATIONAL CONFERENCE ON KINANTHROPOLOGY "Sport and Quality of Life". 2022. ISBN 978-80-280-0107-0.

Základní údaje

• Originální název: Performance analysis of various features and functions settings on the neural network accuracy
• Autoři: BOZDĚCH, Michal (203 Česká republika, garant, domácí).
• Vydání: -13th INTERNATIONAL CONFERENCE ON KINANTHROPOLOGY "Sport and Quality of Life". 2022.

Další údaje

• Originální jazyk: angličtina
• Typ výsledku: Konferenční abstrakt
• Obor: 50902 Social sciences, interdisciplinary
• Stát vydavatele: Česká republika
• Utajení: není předmětem státního či obchodního tajemství
• WWW: 13th INTERNATIONAL CONFERENCE ON KINANTHROPOLOGY "Sport and Quality of Life"
• Kód RIV: RIV/00216224:14510/22:00130440
• Organizační jednotka: Fakulta sportovních studií
• ISBN: 978-80-280-0107-0
• Klíčová slova anglicky: ANN; DNN; inputs and outputs selection and extraction; hidden/output layer activation function; optimization algorithm

Anotace

Artificial intelligence with machine learning (more precisely neural network) is one of the most used types of an algorithm, in sports but also in general. In the past, only wealthier sports clubs could afford to use her potential. With the development of affordable software, the situation changes and every club can use it to their advantage, but there is a problem with how to prepare data and how to set up the algorithm. For these reasons, this study aims to determine whether features or function settings have a greater effect on model accuracy. An initial feature dataset (n = 18,882; 1,929 players, 8 variables) was created from publicly available sources. Each of the 6 feature settings consisted of 96 independent models. A total of 384 models were created, in which their accuracy (in the training and testing phases) and the percentage difference between the training and testing phases were further analysed. No statistically significant differences were found between the accuracy of the function’s settings, but statistically significant differences were confirmed between the features settings, of which14 models with 100% accuracy and 1.00 AUC. From this study, it is clear that feature settings, especially the reduction of the number of outputs, are a more important factor in increasing the accuracy of the artificial intelligence model and that variables Weight, Height, and Age had the highest frequency of occurrences of normalized importance and can therefore be identified as one of the most important features for prediction of final rank.