Statistics Korea(2023) reported that South Korea’s population of people aged 65 or older will exceed 10 million in 2024. It will be a 5.4% increase compared to 2023 and will mark an era in which 20% of the country’s population are senior citizens. Among this population, the mortality rate from COVID-19 is significantly higher. As a result, the elderly are more socially isolated and lonely[1]. Furthermore, with restrictions on contact with family and neighbors, their activity radius has sharply diminished [2]. The COVID-19 pandemic has made it difficult for the elderly to recognize the value of their own lives and live a happy and fulfilling independent life [3]. Constraints on human relationships, quality of life, and social activities have exerted a significant negative impact on socially vulnerable elderly people [4]. In old age, declining physical functions and the increased potential for depression and helplessness are heightened due to the loss of social roles and economic power [5]. In response, leisure and welfare facilities for the elderly, such as senior welfare centers, now offer a variety of leisure services [6]. These programs have been reported to have a positive and meaningful impact on the physical, psychological, and social well-being of the elderly [7].
Given the ongoing practice of social distancing, non-face-to-face delivery systems utilizing online platforms have emerged as alternative service options [8]. In the era of contactless interactions due to COVID-19, online communication methods using IT have become popular. The realm of elderly leisure has seen, various attempts to adapt to this trend [8]. During the pandemic, many in-person social services were suspended and replaced by non-face-to-face services [9]. Social welfare institutions across Korea, for instance, have transitioned their services to non-face-to-face formats using social networking services (social media), messaging platforms, online and videos [10]. This adaptation continues the provision of services despite the challenges posed by COVID-19. With the emergence of new technologies, consumers have become gradually comfortable using the technologies associated with non-face-to-face services [11]. Previously overshadowed by economic motivations such as cost savings through automation and price reductions, consumers have now discovered new psychological desires [12].
However, there are challenges in applying non-face-to-face digital leisure services to the elderly through online platforms. The elderly tend to avoid online non-face-to-face services based on IT media [13]. For these services to be delivered smoothly their recipients must have the capacity to access and use online devices. According to the Digital Information Gap Survey conducted by the Korea Information Society Development Institute(2019), digital literacy among the elderly is the lowest among information-vulnerable groups.
Pirhonen et al. explained the reasons for the low digital utilization capabilities among the elderly[14]. In general, the elderly do not see digitalization as a necessity. They also have less access to expensive smartphones, and fewer opportunities to learn how to use the internet or digital technologies. Smartphone designers have not taken the needs of elderly users into consideration [15]. In addition, services associated with physical activity, and that require actual experience, correction, and feedback are easier to deliver face to face [16]. Nevertheless, the pandemic has led to a decrease in physical activity among the elderly, and an increase in physical and mental health issues. Therefore, developing and disseminating non-face-to-face digital leisure service content for the elderly has become an urgent societal task.
Traditional sports management fields often utilize derivative methods, primarily regression or time series analysis, to predict consumer behavior. Additionally, data-driven analytical approaches are being applied. In big data analytics, various models and algorithms such as decision trees, clustering, artificial neural networks, random forests, and ensemble learning are used for predictive analysis [17]. Among them, artificial neural networks emulate the interconnected structure of the human brain and operate as a supervised learning method, capable of analyzing complex and nonlinear relationships in multivariate data. Artificial neural networks are often regarded as superior to statistical models in estimating consumer behavior accuracy [18].
Understanding the characteristics of elderly participants in sports and identifying the patterns of their participation is a crucial aspect of research on sports behavior Rather than simply using artificial neural networks to predict consumer characteristics, a more effective approach is to classify each group and then segment these groups for prediction, enhancing the hit ratio [19]. Addressing the limitation of artificial neural networks, which may not identify the importance of certain variables, one can use the hit ratio of each group to identify the characteristics of the group with a high predictability for the target variable. Moreover, by proposing artificial neural networks as a complement to cluster analysis, better predictive performance and results can be achieved [21].Therefore, this study has three objectives. The first is to differentiate elderly sports participants by measuring demographic characteristics, frequency of social media usage, and non-face-to-face adaptability, which are criteria for measuring the effects of elderly sports participation in research fields, as variables for digital leisure service participants. The second is to apply artificial neural network models to each group to identify the elderly digital leisure service participation group with the highest likelihood (hit ratio) for the target variable (digital leisure service usage). The third is to analyze the characteristics of the group with the highest target variable and the segmented groups, and to propose strategies for sports participation among the elderly based on these findings(Fig. 1.)
Theoretical Background and Research Scope
Theoretical Background
K-Means Clustering Analysis
The k-Means algorithm widely used in k-Means clustering analysis is an unsupervised learning algorithm that processes large amounts of data without requiring a target feature. The number of clusters, denoted ask, needs to be predetermined, and the appropriate value for k is determined through measures such as the silhouette value [21]. The k-Means clustering process divide a pre-defined dataset into k clusters, and after the k-Means clustering is performed, an additional step is taken to understand the characteristics of each cluster [22].
The k-Means algorithm consists of the following steps:
(1) The random selection of k objects from the dataset as the initial k cluster centers.
(2) The assignment of each data object to the cluster whose center is closest based on Euclidean distance.
(3) The recalculation of the new center for each cluster using the data objects assigned to it.
(4) The repeating of step 2 if convergence criterion is not met.
Artificial Neural Network
Artificial neural networks, a machine learning technique that emulates the human brain [22], consist of nodes in input, hidden, and output layers, as illustrated in Fig. 1. The hidden layer can consist of one or more layers, and the number of nodes in the input and output layers depends on the attributes or representation of the input and output. Once the topology is determined, the model is constructed using the back-propagation algorithm, adjusting the weights between nodes to minimize the error in output values using training data [23]. When input values are provided to nodes in the input layer, the calculated value as the weighted sum to the nodes through arrows connecting lower and upper nodes becomes the output value of the upper node through an activation function [24].
Integration of K-Means and Artificial Neural Network
The performance improvement of the integrated artificial neural network model with k-Means clustering is divided into two forms, as shown in Fig. 2. The first form involves adding labels representing individual clusters resulting from k-Means clustering to the input features of the artificial neural network [25]. In this case, the labels are processed as input values through nodes added to the input layer of the artificial neural network. The second form entails constructing individual artificial neural networks for each cluster resulting from k-Means clustering [26]. In this case, training and testing data for each cluster are prepared, and the reinforcement or improvement of the artificial neural network through k-Means clustering is expressed. This is also referred to as an integrated model of k-Means clustering and artificial neural network.