Tele-rehabilitation is an alternative to the conventional rehabilitation service that helps patients in remote areas to access a service that is practical in terms of logistics and cost, in a controlled environment. It includes the usage of mobile phones or other wireless devices that are applied to rehabilitation exercises. Such applications or software include exercises in the form of virtual games, treatment monitoring based on the rehabilitation progress and data analysis. However, nowadays, physiotherapists use a default profiling setting for patients carrying out rehabilitation, due to lack of information. Medical Interactive Rehabilitation Assistant (MIRA) is a computer-based (virtual reality) rehabilitation platform. The profile setting includes: a level of difficulty, percentage of tolerance and maximum range. To the best of our knowledge, there is a lack of optimization in the parameter values setting of MIRA exergames that could enhance patients' performance. Generally, non-optimal profile setting leads to reduced effectiveness. Therefore, this study aims to develop a method that optimizes the profile setting of each patient according to the estimated (desired) optimal results. The proposed method is developed using unsupervised and supervised machine learning techniques. We use Self-Organizing Map (SOM) to cluster patient records into several distinct clusters. K-fold cross validation is applied to construct the prediction models. Classification And Regression Tree (CART) is utilized to predict the patient's optimal input setting for playing the MIRA games. The combination of these techniques seems to improve the efficiency of the standard (default) way in predicting the optimal settings for exergames. To evaluate the proposed method, we conduct an experiment with data collected from a rehabilitation center. We use three metrics to quantify the quality of the results: R-squared (R²), Mean Absolute Error (MAE) and Root Mean Square Error (RMSE). The results of experimental analysis demonstrate that the proposed method is effective in predicting the adequate parameter setting in MIRA platform. The method has potential to be implemented as an intelligent system for MIRA prediction in healthcare. Moreover, the method could be extended to similar platforms for which data is available to train our method on.

远程康复是传统康复服务的替代方案，可帮助偏远地区的患者在受控环境中获得在物流和成本方面均实用的服务。它包括用于康复锻炼的手机或其他无线设备的使用。此类应用程序或软件包括虚拟游戏形式的练习、基于康复进度的治疗监控和数据分析。然而，如今，由于缺乏信息，物理治疗师对进行康复的患者使用默认的分析设置。医疗互动康复助手（MIRA）是一个基于计算机（虚拟现实）的康复平台。配置文件设置包括：难度级别、容差百分比和最大范围。据我们所知，MIRA 运动游戏的参数值设置缺乏可以提高患者表现的优化。通常，非最佳配置文件设置会导致效率降低。因此，本研究旨在开发一种方法，根据估计的（期望的）最佳结果优化每个患者的配置文件设置。所提出的方法是使用无监督和有监督的机器学习技术开发的。我们使用自组织映射 (SOM) 将患者记录聚类为几个不同的集群。应用 K 折交叉验证来构建预测模型。分类和回归树 (CART) 用于预测患者玩 MIRA 游戏的最佳输入设置。这些技术的组合似乎提高了预测运动游戏最佳设置的标准（默认）方式的效率。为了评估所提出的方法，我们对从康复中心收集的数据进行了实验。我们使用三个指标来量化结果的质量：R 平方（R² )、平均绝对误差 (MAE) 和均方根误差 (RMSE)。实验分析结果表明，所提出的方法可以有效地预测 MIRA 平台中适当的参数设置。该方法具有作为医疗保健中 MIRA 预测的智能系统实施的潜力。此外，该方法可以扩展到类似的平台，这些平台的数据可用于训练我们的方法。