Constructing an artificial intelligence strategy algorithm for the identification of talented rowing athletes

Taiwan’s rowing athletes have performed well during the Asian Games, but their performance in the Olympics has not been adequate. In addition to their hard work and rigorous and effective training, the skill of the athletes is a key factor for achieving good results. In this study, an artificial intelligence (AI) evaluation algorithm is developed to help rowing athletes excel in the sporting events. The AI algorithm uses the analytic hierarchy process to invite experts and scholars in the rowing field to answer a questionnaire. The technique for order performance by similarity to ideal solution is then applied to calculate the ranking of selection indicators, to construct an evaluation model for rowing athletes. The key findings indicate that physicality (or the body structure) is the highest priority among the four main aspects of talent identification; this is followed, in descending order, by specialism, reaction, and psychological elements. The proposed AI strategy was established as the most beneficial decision model and can be used to identify talented rowers in the future.


Introduction
Taiwanese rowing athletes have performed well in the Asian Games. However, their performance in the Olympic Games has not been adequate. Although they were successful in qualifying for the last four Olympics, they did not win any medals. Jiang Qianru, who was ranked 17th in the women's single scull competition in the 2004 Athens Olympics, and Wang Minghui, who was ranked 23rd in the men's single scull in the 2008 Beijing Olympics, remain the best women's and men's rowers for Taiwan in the Olympics. As noted by Tang (1996), rowing in Taiwan can be improved significantly, compared with weightlifting, taekwondo, and archery, wherein the country is a strong medal contender.
Different sports require different physiques and athletic abilities. Therefore, understanding the requirements of a rower's physique can help optimize the process of selection and/or identification of potential rowing athletes. In recent years, although some criticism of sporting talent identification existed (see Roth 2012;Rongen et al. 2018;Till and Baker 2020), much research has been conducted on the selection of athletes across several categories of sports using different methods. Zheng (2017) built a selection model using differences in the physical fitness of excellent male college-level football players for different field positions in China. The model required specific physical fitness tests to be conducted for different positions. Yuan (2017) investigated the body shape and palm dermatoglyphics of the women's volleyball open group in the southern region of Taiwan. The results were compared with those of other relevant studies and are expected to be used for the selection of female volleyball players in the future. A baseball coach adopted the Delphi method to develop a basis for the selection of baseball pitchers at a junior high school (Shen 2017). Cao et al. (2013) referred to the relevant literature on basketball talent selection and collected recommendations of basketball experts to establish an ability assessment model for the selection of professional basketball players. Salimi et al. (2012) used the analytic hierarchy process (AHP) and the technique for order performance by similarity to ideal solution (TOPSIS) to construct a decision-making model for sports venue selection, including nine weighted criteria. Mavi et al. (2012) combined the AHP and TOPSIS algorithms to measure and evaluate the performance of national football team players. Chen et al. (2014) developed a decision-making model for a coaching team to select a starting pitcher scheduling strategy using the AHP and TOPSIS. Furthermore, Nurjaya et al. (2020) adopted the AHP to value certain criteria that help identify talented rowing athletes.
The aforementioned studies indicate that the criteria for identification of sports talent are determined primarily based on the reviews of the relevant literature and the experience of sports experts and scholars. Body shape and physical fitness requirements are the two main criteria included in most studies for the identification of sports talent. Current studies in Taiwan for the identification of rowing athletes are also based on data related to the body shape of high-performing male and female rowing athletes. The data are further compared with that of other highperforming rowing athletes at the international level. An artificial intelligence (AI) strategy algorithm has been used in these studies to discover potential rowing talents; however, the quality of skills required for rowing has rarely been considered (Gong et al. 2012;Huang et al. 2013). Therefore, in this study, rowing literature is referenced, and the skills required by rowing athletes are considered. In addition, based on the AI evaluation algorithm of the AHP and TOPSIS, a model for identifying potential rowing athletes is constructed. Saaty (1980) proposed the AHP model to solve complex decision problems. The AHP is a process that prioritizes the hierarchy and consistency of judgment data provided by a group of decision-makers. The TOPSIS, whose underlying logic is to define the extremes of the ideal and negative ideal solutions, was first proposed by Hwang and Yoon (1981). The ideal solution is the one that maximizes the benefit criteria and minimizes the cost criteria, whereas the negative ideal solution maximizes the cost criteria and minimizes the benefit criteria. In this section, we develop an AI intelligent strategy algorithm for the evaluation of top international rowing athletes. The AI intelligent strategy algorithm comprises the following six steps.

AI intelligent strategy algorithm
Step 1 Establish a hierarchy of experts. The AHP incorporates the evaluation of all decisionmakers into the final decision, without having to elicit their individual ratings on subjective and objective criteria, by pairwise comparisons of the alternatives (Saaty 1990). The AHP can be applied to a diverse array of problems using the following calculation procedure (Delgado-Galvan et al. 2010;Zhang et al. 2010).
Step 2 Development of expert questionnaire for evaluating rowing athletes.
Establish a pairwise comparison matrix A. Let C 1 , C 2 ,..., C n denote the set of evaluation criteria, where a ij represents a quantified judgment on a pair of elements C i and C j . The relative importance of the two criteria is rated using a scale with the values 1, 3, 5, 7, and 9, denoting ''equally important'', ''slightly more important'', ''strongly more important'', ''demonstrably more important'', and ''absolutely more important'', respectively. This yields the following n-by-n matrix A: where a ij = 1 (when i = j); otherwise, a ij = 1 aij , when i,j = 1, 2, …, n.
In addition, 11 experts and/or scholars are involved in answering the questionnaire. Their background is as follows: (1) seven rowing experts, including rowing coaches, each of whom has at least 10 years of coaching experience; (2) three academic experts whose principal area of research is rowing; and (3) one government official who has rowing experience in both coaching and competition and is responsible for the promotion of rowing in a public organization.
Step 4 Calculate the weights of the aspects and criteria. In matrix A, assigning n criteria C 1 , C 2, …,C n a set of numerical weights W 1 , W 2, ..., W n that reflect the recorded judgments is a problem. If A is a consistency matrix, the relation between weights W i and judgment a ij is simply given by Wj If A is a consistency matrix, eigenvector X can be calculated as Saaty (1980) proposed using the consistency index (C.I.) and the consistency ratio (C.R.) to verify the consistency of the comparison matrix. C.I. and C.R. are defined as follows: where R.I. represents the average consistency index over numerous random entries of the same order of reciprocal matrices. If C.R. B 0.1, the estimate is accepted; otherwise, a new comparison matrix is solicited until C.R. B 0.1.
Step 4 Calculation of the TOPSIS decision matrix. The ranking of alternatives in the TOPSIS is based on the relative similarity to the ideal solution, which avoids the situation of both ideal and negative ideal solutions being similar. The calculation processes of the method are as follows (Hwang and Yoon 1981;Behzadian et al. 2012): where A i denotes the possible rowing athletes, i = 1, 2, …, m; X j represents criteria related to alternative performance, j = 1, 2, …, n; and X ij is a crisp value indicating the performance rating of each rowing athlete A i with respect to each criterion X j . Then, the normalized decision matrix R (= [r ij ]) is calculated in Eq. (7).
A set of weights w = (w 1 , w 2 ,…w n ), P n j¼1 w j ¼ 1 from AHP is the accommodated weight (Mavi et al. 2012;Chen et al. 2014;Kashid et al. 2019;Karaköprü and Kabadurmuş 2020). This matrix can be calculated by multiplying each column of R with its associated weight w j . Therefore, the weighted normalized decision matrix V is equal to Eq. (8).
Calculate the distance between the ideal and negative ideal solutions for each rowing athlete using Eqs. (11) and (12).
Step 6 Calculate the comprehensive evaluation value of each rowing athlete.
Calculate the relative closeness to the ideal solution of each rowing athlete.
Step 1 Establish the hierarchy of experts to select rower athletes Step 2 Develop expert questionnaire for evaluating rower athletes Step 3 Calculate weights of the aspects and criteria Step 4 Calculate TOPSIS decision matrix Step 5 Calculate ideal solution and negative ideal solution Step 6 Calculate comprehensive evaluation value for each rower athlete AHP Calculation Process TOPSIS Calculation Process Fig. 1 Construction of the AI evaluation algorithm for rowing athletes Constructing an artificial intelligence strategy algorithm for the identification of talented rowing athletes 1745 where 0 C Ã i 1; that is, a rowing athlete i is closer to A * as C Ã i approaches 1. A set of rowing athletes can be preference ranked according to the descending order of C Ã i .

Results
In this study, eight best rowing athletes were selected from 12 rowing athletes. According to Saaty (1980), the number should not exceed seven so that the consistency of the hierarchy is not affected. Therefore, when not too many elements exist in the rowing athlete level, AHP can only achieve the weight value of each aspect and criteria, and then the TOPSIS method is used to conduct a comprehensive evaluation and ranking of rowing athletes. In this study, based on the AHP and TOPSIS, the rowing athlete AI evaluation algorithm was developed, as shown in Fig. 1. First, the data of the rowing athlete's criteria are measured, and then these data are multiplied by the weights calculated by the AHP. Finally, the TOPSIS calculation process is used to calculate the ranking of the rowing athletes, and the rowing athletes are selected by their ranking. The evaluation steps are as follows: Step 1 Establish a hierarchy of experts to select rowers. Xu (2006) pointed out that selection in sports science is a combination of many sports-related elements. It includes physique, which is related to anatomy, it covers physiological functions and physical fitness elements, which are related to sports physiology, and it covers sports ability, which is related to biomechanics technology and movement. From the field of psychology, mental intelligence and various other psychology-related qualities can also be used as effective assessment criteria to decide whether the athletes have high performance potential. Genetics is also closely related to these sports science elements (Xu 2006). The key factors that influence the selection of rowing athletes through the literature are the four analysis aspects physicality, reaction, specialism, and psychological elements, which in turn give rise to 17 evaluation criteria. The evaluation criteria are five physicality elements (body mass composition, muscle composition, sitting posture, shoulder width, and upper limb length), four reaction elements (explosive force, coordination, speed, and muscle    Constructing an artificial intelligence strategy algorithm for the identification of talented rowing athletes 1747 endurance), four specialism elements (flexibility, dynamometer results, pull strength, kick strength), and four psychological factors (stress resistance, concentration, selfconfidence, and goal setting). According to the literature, the four aspects and 17 selection criteria for evaluating the selection of rowing athletes were identified and analyzed using the AHP method. The method is divided into three levels: The selection of the first layer indicates the rowing athlete, the second layer indicates the four aspects for each rowing athlete, and the third layer is the 17 criteria and mathematical code (Huang 2019; see Table 1, and the AHP architecture diagram in Fig. 2).
Step 2 Development of an expert questionnaire for evaluating rowing athletes.
According to Step 1, a pairwise comparison is determined, and a questionnaire for the experts is developed to evaluate all the aspects and criteria. Taking the evaluation aspects as an example, the questionnaire of the expert pairwise comparison is shown in Table 2.
Step 3 Calculate the weights of the aspects and criteria.
Equations (2) and (3) are used to calculate the weights of the aspects and criteria (see Table 3).
Step 4 Calculation of the TOPSIS decision matrix. The weights of the aspects and criteria are multiplied by the measurement values of each rowing athlete, which are the weighted measurement values of each rowing athlete, by using Eqs. (6)-(8), and this value is used as the initial calculated value of the TOPSIS's decision matrix (Table 4). Moreover, the background of the selected rowing athletes is as follows: sex = male rowers; average age = 16.12 ± 1.05 years; sporting achievement = the trained rowers were recruited from the top eight in the national high school competition team that volunteered to participate in this study; participants with C 3 years of rower experience; they were training in rowing for 16 h at least per week.
Step 5 Calculate the ideal and the negative ideal solutions.
Equations (9)-(12) are used to calculate the ideal and the negative ideal solutions (Tables 5 and 6).
Step 6 Calculate the comprehensive evaluation value of each rowing athlete.

Conclusion and discussion
This study contributes to the establishment of an AI intelligent strategy algorithm for selecting rowing athletes. First, four main aspects and 17 criteria were established by using the AHP to collect the recommendations of experts. The weights and importance of the main aspects and criteria were then evaluated. The TOPSIS was used to compute the evaluation index, which was then used to select the rowers. In the future, an empirical analysis, such as how well each rowing athlete does in competitions, will be used to verify the reliability of the model for selecting rowing athletes. Moreover, the statistical data of more elite rowing athletes will be included into the analysis for fine-tuning. Furthermore, other sports can construct their own athlete identification model by following the strategy elucidated in the current study. As Johnston et al. (2018) mentioned, there is a need for diverse research on sports talent identification. Finally, different decision-making strategies can be applied and compared with the strategy applied in this study.
Author's contributions Jing-Wei Liu contributed to writing-original draft, conceptualization, constructing experimental model, and methodology. Sheng-Hsiang Chen contributed to methodology, validation, and writing-review and editing. Che-Hsiu Chen contributed to data curation. Tsung-Han Huang contributed to project administration and data curation.
Funding The authors did not receive support from any organization for the submitted work.
Availability of data and materials Not applicable.
Code availability Not applicable.

Declarations
Conflict of interest The authors declare that they have no obvious competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
Ethics approval This article does not present any studies with human participants or animals performed by any of the authors.