Decision support system play an important role in identifying, assimilating the nature of complex Data [19]. Data can be categorized and selected based on their priority and preferences to enhance the process and operation of various tasks in large organizations [20]. Fister et al. [21] have conducted research on the implementation of various algorithms based on computational intelligence for the enhancements in sports activities. The point of the proposed article is to show that nature-propelled algorithms are additionally helpful inside the space of the game, specifically for acquiring protected and viable preparation plans focusing on different parts of the performance. At last, a reconciliation of the outcomes got for the various region of sports preparation ought to empower the making of a counterfeit fitness coach. This could be useful for competitors who cannot manage the cost of coaches due to high expenses. The article focused on the implementation of virtual reality for the development of sports training architecture along with the creation of motion capture data procedure grounded on behavior string. To evaluate the efficiency of the proposed architecture, the learning of tennis by the school student from the beginning was considered as an experimental object. It was found that the student was greatly impressed by the usage of the modern technique and his interest in tennis learning was enhanced. The study showed that based on virtual reality, a very effective and efficient training atmosphere can be developed for the training of new learners [22]. The aggression shown by the players during sports can result in some serious situations and can affect the quality of sports. Deng et al. [23] have presented a system grounded on the swarm intelligence and the employment of the Internet of Things for the prediction of aggressive behavior from the players. After the recognition of various emotions based on the proposed model, the related personnel can act in time and resolve the issue very conveniently. The main emotions predicted by the developed system are anxiety, surprise, and sadness. The experimental data revealed that the system can achieve an accuracy of about 80%. Lu et al. [24] have developed a cost-effective system for the training and talent selection grounded on artificial intelligence and the Internet of Things (AI + IoT). The proposed system can work on very less computational assets and can be employed in foot-driven sports. The framework is composed of wireless wearable sensing devices (WWSDs), mobile application, and data processing nodes. The evaluation revealed that the architecture can efficiently identify the motion and skills of the players with an accuracy rate of about 93%. It is more effective than the existing approaches of monitoring motion based on the analysis of video. Wang and Li [25] have conducted research to help the athletes in carrying out their training with the help of videos without any constraints of space and time. The proposed model employed a machine learning grounded algorithm named cut vertices spanning tree for the distribute multicast training and dynamically adjusting the number of layers of sports video streaming. The experimental data show that the system can perform better in terms of quality of experience, packet loss rate, and link utilization. The system is very efficient in the training of players by providing them with videos very conveniently.
The behavior of the human body can be studied very precisely by the employment of artificial intelligence grounded techniques. The study proposed a paradigm for the diagnosing and analysis of the features of the human body after training by the implementation of a detection framework in integration with a convolutional neural network (CNN). For the analysis of the motion of the human body, the system adjusts neurons according to the map of the skeleton. The output data revealed that CNN is very productive in the recognition of the fatigue level of a body after training. The overall performance of the system is very impressive and is very robust and applicable [26]. Chen and Hu [27] have developed a model for the precise identification of the motion of a player based on deep learning neural network. At first, the movement was decomposed into a series of movements and then thoroughly judged by the usage of a mobile neural network (MNN) inference engine. The proposed system was compared with the existing approaches and the results revealed that the system is very efficient and effective than these approaches. The system can achieve high performance and accessibility and enhance the accuracy of motion recognition in sports.
3.1 Extracted features and selection
The volume of data that is available on the internet has expanded dramatically in recent years. As a result, machine learning approaches struggle to comply with the enormous amount of computational variables, offering an exciting challenge for experts. Feature extraction is required in order to utilize advanced analytics appropriately. Feature selection is a significant approach in data preparation, and it has become an essential element of the learning algorithm. In data mining research, it is sometimes referred to as variable subset selection, attribute selection, or alternates selection [28]. Feature selection is a data processing mechanism that make data to be presentable in various structures[29]. A feature is a unique, quantifiable aspect of the process under observation. Classification may be carried out using neural network model from a collection of features. In recent years, the scope of features employed in deep learning or supervised classification applications has risen from up to hundreds of independent factors. Several strategies have been developed to reduce unnecessary and superfluous variables that are a burden on difficult assignments. Feature selection aids in data assimilation, reduces processing requirements, mitigates the influence of the computational complexity, and improves predictive accuracy. In this study, we examine some of the strategies identified in the literature that employ specific measures to identify a selection of variables (features) that increase overall accuracy rate. Selection of relevant features from data pattern is carried out to make the processing faster and more reliable by removing redundant and irrelevant features [30, 31]. The features are extracted from literature reviews and are classified into relevant and irrelevant features. The relevant features are then fed into the proposed system to enhance the learning process of machine and to provide accurate results.
Table 1
Features Analyzed in literature
| Citations | Features | Citations | Features |
| [1] | Wrong actions, accuracy, action recognition, time, training | [13] | Players’ evaluation, performance, match statistics, team sports |
| [2] | Training efficiency, action recognition, classification | [14] | Assisted judgement, players’ performance, performance evaluation, unbiased decisions |
| [3] | Environmental brightness, recognition, players’ poses, human targets, accuracy | [15] | Performance measurement, reliability, correlation analysis, efficient decision making, sports arena |
| [4] | Sports skills, sportsmen’s motivation, action recognition, accuracy | [16] | Players’ ranking, playing time, decision making |
| [5] | Site selection, sports facilities, population distribution, accuracy | [17] | Athlete monitoring, data visualization, athlete performance, coach assistant |
| [6] | Sports training, personalized learning, effective sports platform | [18] | Sports venues, decision making, efficiency, information management |
| [7] | Outdoor sports, weather conditions, field conditions, accurate decision | [21] | Simplicity, efficiency, performance |
| [8] | Player motions, decision making, sports strategy, efficiency, visualization | [22] | Sports event, students’ guidance, students’ training, improvement |
| [9] | Sports marketing, customer information, marketing plans, sports managers | [23] | Emotions recognition, athletes’ emotion, accuracy, anxiety, sadness |
| [10] | Exercise suggestions, non-standard actions, actions recognition, constructive suggestion | [24] | Motion tracking, players’ skills, talent selection |
| [11] | Students’ guidance, students’ efficiency, physical fitness, time-saving | [25] | Improves quality, players’ training, video analysis |
| [12] | Physical education, teaching resources allocation, athletic sports, accurate decisions | [26] | Body behavior, sports training, recognition model, feasibility, accuracy |
| | | [27] | Action recognition, movement details, accuracy, accessibility |
Table 2
| S. No. | Features |
| S1 | Players’ evaluation |
| S2 | Action recognition |
| S3 | Physical fitness |
| S4 | Sports venues |
| S5 | Sports training |
| S6 | Athletes’ emotions |
| S7 | Field conditions |
| S8 | Motion tracking |
3.2. Ant Colony Optimization (ACO)
Optimization challenges are extremely important in both the technological and research domains. Network selection, routing path selection, crowd selection, task assignment [32, 33], device selection [34], shape optimization, and gene identification [35] are some examples of practical optimization challenges. ACO is a feasible approach for improving prediction. Combinatorial optimization is inspired by ant colonies in nature. The fundamental motivator for ACO is the foraging actions of ants [36]. Ants will explore the region surrounding their colony at random in search of food. When an ant finds a food source, it assesses the quantity and taste of the food and brings some of it to the nest. The ant leaves a chemical pheromone trail on the earth on its way back. The amount of pheromone deposited, which varies according to the quantity and quality of the food, will guide other ants to the food source. As previously proven, indirect communication among ants via pheromone trails assists them in determining the quickest routes between their colony and food sources. This natural ant colony feature is employed in artificial ant colonies to overcome CO problems. ACO algorithms use this feature of actual ant colonies to address combinatorial optimization issues [37]. To use ACO, the problem is turned into the challenge of determining the optimum route on weighted graph (node, edges). The artificial ants create solutions iteratively by travelling over the graph. The process is probabilistic and is influenced by a chemical modelling, which is a collection of traits associated with graph elements (either nodes or edges) whose contents are adjusted at execution by the ants [38].
3.3. ACO Based Feature Selection for action Recognition
ACO based feature selection model can be utilized to solve the problem of action recognition of sport mans to enhance their performance in national and international competitions. The flow chart is presented in Fig. 1, the process of athletes feature selection starts with the ant’s generation from their nest.
After initialization the movement i.e. the traversal of ants starts that move across random paths to find their source node also termed as food node. In the traversal process the ants excrete a chemical termed as pheromone that is used as a signal for ants to track their own path or to assist other ants to find companions route. In this process the ants select various nodes. These nodes are then evaluated based on pheromones value. The path that have more pheromones are declared as best paths as more ants move across these paths and the associated nodes are selected as they are appropriate nodes. A subset of original features is formed as a result of this traversal that is used as training set for enhancing learning processes. The ACO in this study can be mapped in three important stages. The graphical method represented in Fig. 2 in which nodes are linked with edges.
The ants traverse over these edges in search of finding best features among a group of features. Nodes represent various features of sport man. The ants select the best features to enhance the athletic activity in various competitions. A set of most important features S1, S2, S3, S4, S5, S6, S7 are collected from literature. Pheromone intensity and heuristic desirability is a second important stage in ACO Ants are allowed to move freely on the graph in their first iteration. The movement is accompanied with the excretion of pheromones on edges represented in Fig. 3 and their probability in Fig. 4.
The follower’s ants use transition table i.e., pheromone signaling and heuristic information strategy to select a path and a corresponding node. If the ants find its goal the process end otherwise iterates for another cycles. Ants on S1 evaluate which node (feature) should be picked, the evaluation is carried out based on the high volume of pheromones on connected paths. using the formula (1) the decision of selecting a node is carried out:
P (Edge) = P (Pheromones (Xi) ηi) / ∑ (P (Pheromone (Xi)) ηi) …. (1)
The Edges of ant’s traversal is represented as Xi, ηi is the heuristic desirability. If heuristic desirability value is high the edge is selected otherwise neglected. The pheromone value is a motivation of ant traversal i.e. ants selects paths with high pheromones value.
After reaching termination goal the ants end its traversal and provides an output subset, that may not be equal as original set as only optimal (i.e., high pheromones value) nodes is selected represented in Fig. 5. This subset is then used for enhancing the capability of athletes. If the ants do not provide optimum results in an iteration, the ants repeat the traversal this stage is termed as pheromone update process (Fig. 1). The pheromone value is incremented using the Eq. (2).
τι (τ+1) = (1−ρ).τι (τ) + ρ . Δτι(τ)....(2)
τι is the remaining pheromones on paths that are not evaporated yet, pheromone decay that is evaporated is represented by ρ symbol the value ranges in between 0–1. Δτι is pheromone modification or increment for subsequent iteration, the ants on leaves more pheromones on optimal paths and as a results provides best nodes (features) of athletes. Best paths with the corresponding optimal features are presented in Fig. 5.