Academic spaces are an environment that promotes student performance not only because of the quality of its equipment, but also because of its ambient comfort conditions, which can be controlled by means of actuators that receive data from sensors. Something similar can be said about other environments, such as home, business, or industry environment. However, sensor devices can cause faults or inaccurate readings in a timely manner, affecting control mechanisms. The mutual relationship between ambient variables can be a source of knowledge to predict a variable in case a sensor fails. Moreover, the relationship between these variables and the occupation of spaces by students over time also contains an adequate knowledge of the context for prediction. In this article we propose to predict ambient variables by means of recommendation systems based on collaborative filtering, which are fed with data from sensors over time in different academic rooms. For this purpose, we applied two different algorithms: Probabilistic Matrix Factorization and Bayesian Non-negative Matrix Factorization. The accuracy of the algorithms when comparing actual and predicted values and the performance comparison between the two collaborative filtering implementations lead us to propose Probabilistic Matrix Factorization as a good approach for supporting ambient control systems.

学术空间不仅可以提高设备质量，而且还可以提高周围环境的舒适度，从而可以提高学生的学习水平。环境舒适度可以通过从传感器接收数据的执行器来控制。关于其他环境，例如家庭，企业或行业环境，可以说类似的话。但是，传感器设备可能会及时导致故障或读数不正确，从而影响控制机制。在传感器发生故障的情况下，环境变量之间的相互关系可以成为预测变量的知识来源。此外，这些变量与学生随时间推移所占空间之间的关系也包含对预测环境的充分了解。在本文中，我们建议通过基于协同过滤的推荐系统来预测环境变量，这些系统将随时间推移从不同学术教室的传感器数据中获取数据。为此，我们应用了两种不同的算法：概率矩阵分解和贝叶斯非负矩阵分解。当比较实际值和预测值时算法的准确性以及两个协作过滤实现之间的性能比较使我们提出概率矩阵分解作为支持环境控制系统的一种好方法。