Machine Learning for the Prediction of Physical Fitness Levels

• Main Authors: Chieh-Ming Yang, 楊哲旻
• Other Authors: Mei-Juan Chen
• Format: Others
• Language: zh-TW
• Published: 2018
• Online Access: http://ndltd.ncl.edu.tw/handle/96xz6j

碩士 === 國立東華大學 === 電機工程學系 === 106 === Science and technology bring people convenience and change our life habits, leading to a reduction of the daily volume of exercise. In the modern world, it is necessary to know the factors related to fitness in order to improve the physical fitness of general population. Professional military personnel have to receive regular and tense exercise training. We take a large sample of the military members for machine learning for the baseline bio-psychological factors to predict the performance of physical fitness and the most- related factors. This research uses six machine learning techniques including logistic regression, decision tree, random forest, gradient boosting regression tree, support vector machine and multilayer perceptron to predict the level of physical fitness of the military members. The physical fitness level is evaluated by 3000-meter running, 2-minute push-up and 2-minute sit-up. The pre-processed procedures include sampling, feature reduction and feature selection. The performance of each machine learning technique is evaluated by accuracy, confusion matrix, 10-fold cross validation and area under receiver operating characteristic curve (AUC). The results show that the predictive performances for the three exercise tests(3000-meter running, 2-minute push-up and 2-minute sit-up) after machine learning are higher in the lower level of group (accuracy: 0.924, 0.873 and 0.916, respectively; cross-validation: 0.894, 0.847 and 0.900, respectively; AUC: 0.914, 0.827 and 0.928, respectively). In the machine learning techniques, we find that the predictive performance of multilayer perceptron is better than those of other models.