Indoor Environmental Quality Management in Learning Spaces through Predictive Models of Students' Physical and Emotional Well-Being

This repository contains the resources developed for predicting students' well-being in educational buildings based on environmental conditions and students' feedback. This repository includes two pre-trained machine learning models designed to estimate the physical and emotional well-being of students based on both objective classroom conditions and subjective student perceptions of comfort and well-being. The first model, labeled rf_bf.pkl, predicts the physical well-being of students, while the second model, labeled rf_ba.pkl, predicts their emotional well-being. These models take into account a variety of factors, including the environmental conditions inside and outside the classroom, as well as the condition of the physical infrastructure. The models were developed using data collected over a period of five months from classrooms monitored at the Universidad Politécnica de Madrid, ensuring that they reflect a wide range of real-world classroom settings. Additionally, this repository contains a data scaler, provided as the minmax_scaler.pkl file, which is used to preprocess the input data before it is fed into the models. This scaler ensures that the input data matches the feature distribution used during the training of the models, maintaining the accuracy and integrity of the predictions. To facilitate the use of these models, a graphical user interface has been developed and implemented as a PowerBI dashboard, provided in the Well-Being Prediction.pbix file. This user-friendly interface allows users to input different classroom conditions in an interactive manner. Based on these inputs, the interface then generates predictions regarding the physical and emotional well-being of students. In order to ensure the interface functions correctly, a comprehensive setup guide is included, detailed in the Prediction Dashboard Setup Guide.pdf file. This guide outlines the necessary steps for configuring and installing the required software and tools, ensuring that users can easily access and utilize the interface to obtain accurate well-being predictions. Additionally, a requirements.txt file is provided, listing all the dependencies needed for the correct operation of the dashboard. A more detailed explanation of the requirements and the setup process can be found in the guide. The repository also contains the file RelatedWorkAnalysis.xlsx, which includes the analysis of the related work conducted as part of this study. The synthesis of this analysis is presented in the paper, while the full comparative dataset is provided in this repository to support transparency, consultation, and reuse by other researchers. It is important to note that the training data used for the prediction models is in Spanish. Therefore, the values you need to select during the inference process are also in Spanish. Below is the translation of the values used in the models: Sí = YesNo = NoParcialmente = PartiallyO (oeste) = O (west)SE (sureste) = SE (southeast) When performing the inference process and asked to select these values, make sure to choose the corresponding options in Spanish (or use the provided translation depending on the system language). Ensure that the selections are consistent with the expected values for accurate predictions.