运动科学综述阅读(5): Performance Analysis

Performance Analysis in sports

运动表现分析可以应用于评价运动员、团体的表现。能够帮助运动员/教练作出正确的调整方向/目标。
然而，庞大的数据中可能只有一少部分能够使用。所以，利用一些优化算法确定Feature Selection也是主流的做法之一。

下一章主要包括Data Mining中Video Analysis部分，与本章重复的部分没有再赘述。笔者前两天准备了gre考试+考的时候就有点发烧，考完直接睡到第二天，摸鱼了两天，哈哈

Fister Jr et al., 2015

detailed survey on how computational intelligence can influence sports
focus on achieving effective training plans, targeting aspects of performance by computational intelligence algorithms, including machine learning and fuzzy concepts

Cordes and Olfman(2016)

select suitable features
GA + feature subset selection(FSS)
high computational overhead

Fister Jr et al. (2017)

Post hoc analysis
marathon
loss between the predicted and the achieved results by an athlete after the performance, emphasis on marathon
deficit time between predicted time and real cost time
DE sensors in sports watches -> route, GPS, constraint satisfaction problem(CSP)
43 variable->intermediate pace deficit for each kilometers(assumed 43 km in total)
standard differential mutation, uniform crossover

Fister et al. (2017)

half IRONMAN
preferred time using PSO
5 elements:
●swiming, transition 1
●cycling, transition 2
●runnig
boundary constraints for each criterion: based on past data
fitness function: Pearson correlation between the time difference in swimming and cycling, and the time difference in cycling and running
PSO searched for a solution that mostly fit with the correlation of the result archived->able to predict the time based on historical data

Fister et al. (2020)

include heart rate data
DE
special concerns were paid when setting boundaries, this time for the heart rates
identified some facts: like ‘in which paces the runner can speed up and slow down to have optimum performance’ etc.

Mlakar and Luˇstrek (2017)

Tennis
multi-objective optimization&machine learning
●new data capturing device related to performance measures Catapult Sports
●ML-> feature capture NSGA-II
●multi-objective optimization->detect the start and end of a rally( defined by some rules). 2 objects: Classification error and the number of missed shots
●pareto front for graphical interpretation

one of the initial achievements in evaluating player performance in the game of tennis

Geng and Hu(2020)

basketball
GP
tree based GP with 16 features. Implemented addition, subtraction, multiplication and protected division
●For one season, ranking error for all the teams were evaluated
●the error of all teams in all seasons is considered as the fitness
crossover/mutation: sub tree replacement
parameters of the GP: trial and error approach

Marcelino et al. (2020)

football
how collective behavior of payers influences the success of the game
no specific meta-heuristic. Adopt the concepts in natural collective behaviors such as bird flocks and fish schools which supposed to be the inspirational aids for algorithms like PSO and AFSO

Romero et al. (2021)

PSO
find MVP in a handball game
problem: find MVP from m players and n performance criteria
weight of feature set-> rank
among the three approaches, the PSO was unable to achieve the expected higher outcomes for the weights compared to the other two(Geng and Hu (2020) and Marcelino et al. (2020))

Jana and Hemalatha (2021)

players performance in football
optimizing a multi linear regression model
KNN-> identify top players for a given attribute or skill

Lee et al. (2019)

baseball batter evaluation model
factors such as run value, contribution score and ball consumption.-> weighted in an optimized manner
GA
language barriers:(