Machine Learning Techniques in Indoor Environmental Quality Assessment

Abstract

This chapter provides a comprehensive exploration of the evolving role of machine learning in Indoor Environmental Quality (IEQ) assessment. As urban living spaces become increasingly enclosed, the importance of maintaining optimal IEQ for human health and well-being has surged. Traditional methods for IEQ assessment, while effective, often fail to provide real-time monitoring and control. This gap is increasingly being addressed by the integration of machine learning techniques, allowing for enhanced predictive modeling, real-time optimization, and robust anomaly detection. The chapter delves into a comparative analysis of various machine learning techniques including supervised, unsupervised, and reinforcement learning, demonstrating their unique benefits in IEQ assessment. Practical implementations of these techniques in residential, commercial, and specialized environments are further illustrated through detailed case studies. The chapter also addresses the existing challenges in implementing machine learning for IEQ assessment and provides an outlook on future trends and potential research directions. The comprehensive review offered in this chapter encourages continued innovation and research in leveraging machine learning. for more efficient and effective IEQ assessment.

1. Introduction

This chapter delves into the integration of Indoor Environmental Quality (IEQ) assessment with Machine Learning (ML), emphasizing the critical role of optimal IEQ in urban living spaces. It explores the impact of IEQ on occupant health, comfort, and productivity [1], and examines how ML can revolutionize IEQ assessment, offering significant advantages for building management, urban planning, and public health sectors.

IEQ encompasses the conditions within buildings that affect human health and comfort, including air quality, thermal comfort, acoustical comfort, and visual comfort [2]. These elements, as illustrated in Figure 1, significantly influence occupant well-being and productivity [4]. The scope of IEQ spans various building types, from residential to commercial and institutional spaces like schools and hospitals [5]. Monitoring of IEQ involves diverse methods, ranging from manual controls to automated systems, and adheres to standards set by organizations like the EPA and ASHRAE [6].

Figure 1.

Factors Influencing IEQ [3].

Accurate and efficient assessment of IEQ is essential for maintaining healthy indoor environments. It aids in optimizing HVAC systems for energy efficiency and comfort [7] and addresses health concerns such as Sick Building Syndrome (SBS) and Building-Related Illnesses (BRI) [8]. Traditional IEQ assessment methods often fall short in real-time monitoring and managing the dynamic nature of indoor environments [9].

Machine learning, a transformative branch of artificial intelligence, is increasingly applied in IEQ assessment. ML offers advanced capabilities in data processing, predictive modeling, pattern recognition, anomaly detection, and real-time adaptive control [10]. Through various techniques, such as supervised, unsupervised, and reinforcement learning, ML enhances the accuracy and efficiency of IEQ assessments [11].

IEQ is a complex, multifaceted concept integral to occupant well-being. It spans a wide range of parameters, from physical to biochemical and psychological aspects [12]. The trend towards energy-efficient building designs has intensified the need to focus on maintaining IEQ [7], which includes factors like air quality, lighting, thermal conditions, and acoustics, as well as emerging considerations such as ergonomics, esthetics, and electromagnetic field quality, as defined by entities like ASHRAE, the EPA, and WHO [13, 14, 15].

Each aspect of IEQ, critically impacting human health and comfort, is depicted in Figure 1. For instance, poor air quality can lead to a range of health issues, from allergies to severe conditions like cancer [1]. Inadequate thermal conditions can diminish comfort and reduce productivity [2], while neglect in acoustical and visual comfort can result in stress, cognitive impairments, and long-term mental health issues [16, 17].

To illustrate further, air quality is influenced by various pollutants, both internal and external to the building, with effects ranging from minor discomfort to serious conditions like asthma and cancer [18]. Thermal comfort, a subjective but crucial metric, hinges on multiple environmental and personal factors [2]. Acoustical comfort, dependent on sound characteristics, is vital for mental well-being [16]. Visual comfort, shaped by lighting and visual elements, affects everything from eye strain to mood and circadian rhythms [17].

The integration of ML into IEQ assessment represents a significant advancement. ML’s ability to analyze complex datasets and discern patterns enables a more nuanced understanding and control of indoor environments. This chapter aims to explore the applications of ML in enhancing various facets of IEQ, from air quality analysis to thermal comfort prediction. The integration of ML in IEQ assessment not only promises improved real-time monitoring and proactive management but also opens doors to personalized environmental controls tailored to individual needs and health requirements.

As we delve into the specifics of applying ML to IEQ, the chapter will address both the technical advancements and the challenges, including ethical considerations. Our goal is to provide a comprehensive overview of how ML can be leveraged to improve indoor environmental quality, ultimately contributing to the enhanced health, comfort, and well-being of occupants in urban living spaces.

2. IEQ assessment: Traditional methods and their limitations

The assessment of Indoor Environmental Quality (IEQ) plays a pivotal role in ensuring environments that support the health, comfort, and productivity of building occupants. While traditional methods of IEQ assessment have been the cornerstone of this field, they exhibit significant limitations, impacting their overall effectiveness and reliability. In this section, we will expand upon these traditional methods, underscoring their constraints and paving the way for the introduction of machine learning techniques as a more robust solution.

2.1 Traditional assessment methods

2.1.1 Spot measurements

• Description: This method involves taking discrete measurements of various IEQ parameters such as temperature, humidity, carbon dioxide levels, particulate matter, and light intensity at a specific moment using specialized instruments [19].

• Advantages: The primary benefits include simplicity in execution and the immediate availability of results, making it a convenient option for quick assessments.

• Limitations: However, spot measurements may not accurately represent ongoing exposure conditions, as they overlook temporal and spatial variability. The accuracy of these measurements is also heavily dependent on the calibration and precision of the instruments used [20].

2.1.2 Long-term monitoring

• Description: This approach utilizes sensors or data loggers to continuously record IEQ parameters over extended periods. It is designed to capture daily and seasonal variations, providing a more comprehensive view of the indoor environment [21].

• Advantages: Offers a broader and more detailed perspective on IEQ, capturing long-term trends and fluctuations.

• Limitations: The challenge lies in the significant resources required for data management and analysis. Additionally, there is a potential for data errors and inconsistencies due to sensor calibration issues or environmental interference [22].

2.1.3 Occupant surveys

• Description: This method involves collecting subjective responses from occupants about their perceptions and satisfaction with the indoor environment [23].

• Advantages: Occupant surveys can uncover issues that might be overlooked by physical measurements, providing valuable insights into the human aspect of IEQ.

• Limitations: However, these surveys are prone to response biases and may not accurately reflect the actual IEQ conditions. The qualitative nature of this data also makes it challenging to quantify or systematically analyze [24].

2.1.4 Physical inspections

• Description: Physical inspections involve thorough examinations of the building by professionals to identify visible or detectable issues such as water leaks, mold growth, or insufficient ventilation [25].

• Advantages: They are particularly effective in identifying overt and tangible problems within a building.

• Limitations: The limitation of physical inspections lies in their inability to detect latent or intermittent problems. They also require significant expertise and can be time-consuming and labor-intensive [26].

2.2 Limitations and the need for advanced methods

The traditional methods discussed above share a common drawback: their limited ability to provide real-time, continuous data. They often fail to capture the dynamic and interactive nature of various IEQ parameters, leading to an incomplete or skewed understanding of the indoor environment. Furthermore, these methods generally lack predictive capabilities, which are essential for proactive IEQ management. They can also be resource-intensive, both in terms of time and financial investment [27].

2.3 Emerging advancements and the role of machine learning

In response to these limitations, emerging advancements in sensor technology and data analytics, particularly machine learning (ML), present promising solutions. Machine learning techniques, with their capability to process vast amounts of data, enable continuous monitoring, real-time analysis, and predictive modeling of IEQ parameters. This paradigm shift aims not only to enhance the accuracy and comprehensiveness of IEQ assessments but also to streamline the process, reducing the demand on resources. The integration of ML in IEQ assessment represents a significant step towards more intelligent, adaptive, and occupant-centered building management.

However, it is important to note that the full-scale adoption of these advanced technologies is still in progress. Further research and development are necessary to establish their effectiveness, ease of use, and integration into existing building management systems. As the field of IEQ assessment evolves, the potential of machine learning to transform this domain is immense, offering a pathway towards more sustainable, healthy, and productive indoor environments [27, 28].

3. An overview of machine learning techniques

Machine learning, as a pivotal subfield of artificial intelligence, has seen notable evolution since its inception. This discipline, coined by Arthur Samuel in 1959, initially revolved around the idea of giving “computers the ability to learn without being explicitly programmed” [29]. Samuel’s work with checkers-playing programs laid the groundwork for subsequent progress in this field. The following years saw pioneering efforts from researchers like Rosenblatt who, in 1958, introduced the concept of the perceptron, an early neural network that could classify linearly separable patterns [30]. At the same time, Nilsson and Minsky’s explorations in learning machines added depth to the understanding of how computational models could replicate cognitive processes [31].

The 1970s and 1980s introduced the concepts of decision trees [32], a fundamental machine learning approach which paved the way for more sophisticated algorithms like Random Forests. This era also witnessed the emergence of the kernel methods, which includes the Support Vector Machine (SVM) - a powerful classification tool that marked a significant shift in the direction of machine learning [33]. The advent of the Backpropagation algorithm in the 1980s sparked a renewed interest in neural networks [34]. However, the full potential of this method wasn’t realized until recent years, due to the limitation in computational capabilities and lack of large-scale, labeled datasets.

The 1990s saw a move towards data-driven approaches, prompted by the increasing availability of digital data. The introduction of ensemble methods such as boosting and bagging brought forth a new phase in machine learning, improving model robustness and predictive power [35]. The turn of the millennium marked the era of ‘big data’, pushing the boundaries of machine learning further. The development of more sophisticated algorithms, like deep learning, could now be propelled by the explosion of data and advancements in computational capabilities [36]. The seminal AlexNet model, developed by Krizhevsky, Sutskever, and Hinton, for the ImageNet competition in 2012, demonstrated the incredible potential of deep learning in practice, revolutionizing the field and driving a resurgence in neural networks [37].

Today, machine learning has permeated various sectors, revolutionizing healthcare, finance, environmental science, and many more fields. From a theoretical concept to a practical tool, machine learning has become an integral part of scientific research and technological development.

3.1 Fundamentals of machine learning

In this subsection, we will delve into the core concepts that form the foundation of machine learning, encompassing its definition, the nature of the data it utilizes, the problems it addresses, and the metrics used to evaluate its performance. Machine learning, a subset of artificial intelligence, enables computers to learn autonomously by training mathematical models on data, thereby going beyond the explicit programming used in traditional computing [38]. Unlike classical programming that relies on pre-defined rules, machine learning algorithms identify patterns within data, offering a unique capacity for experiential learning.

Data is essential for machine learning performance. It can vary in type—numerical, categorical, or a blend—and often requires preprocessing steps such as data cleaning and missing data imputation to ensure model quality [38]. Machine learning can address multiple problem types, including but not limited to regression, classification, clustering, anomaly detection, and recommendation systems [39].

Evaluating the effectiveness of a machine learning model involves various metrics. In classification tasks, commonly used metrics include accuracy, precision, recall, F1-score, and ROC-AUC [40]. For regression tasks, mean absolute error, mean squared error, and R-square are commonly applied [38]. The choice of evaluation metric is contingent on the specific problem and objectives, making a thorough understanding of these elements vital for developing robust machine learning models.

3.2 Relevance of machine learning in environmental data analysis

Environmental systems are inherently complex, characterized by high variability and unpredictability. Traditional statistical models often struggle to capture the nuances of these systems due to their linear nature and the assumption of independence among predictors. Machine learning, on the other hand, offers the flexibility to model non-linear relationships and consider interaction effects among multiple variables, making it adept at capturing the complexity inherent in environmental systems [41].

Machine learning’s strength lies not just in its ability to model complex systems, but also in its ability to handle unstructured data. Unlike traditional models that require structured tabular data, machine learning algorithms can handle a variety of data types, including text, images, and sound. For instance, images from remote sensing can be used in land use classification or change detection tasks using convolutional neural networks, a type of deep learning algorithm [42]. Text from social media posts can be analyzed using natural language processing techniques to gauge public sentiment towards environmental issues or to detect early signs of natural disasters [43].

Time-series analysis is another critical aspect of environmental data analysis. Environmental data is often collected over time, leading to a temporal sequence of observations. Machine learning plays a crucial role in the analysis of these time-series data. Recurrent neural networks (RNN) and Long Short Term Memory (LSTM) networks, types of deep learning algorithms, have shown significant promise in modeling temporal dependencies in environmental data, such as weather forecasting and prediction of air quality indices [44].

Lastly, the importance of feature engineering cannot be understated in the context of environmental data analysis. Feature engineering is the process of creating new features or modifying existing ones to improve model performance. While machine learning algorithms can automatically learn features in some cases (particularly in deep learning), domain knowledge can often guide more effective feature creation. For instance, in the case of predicting rainfall, features such as the season, geographical location, historical weather patterns, and other atmospheric conditions might be crucial. Incorporating such features in the model could potentially improve the model’s predictive performance [38, 45].

3.3 Supervised, unsupervised, and reinforcement learning: a comparative analysis

This subsection aims to delve deeper into the nuances of supervised, unsupervised, and reinforcement learning, providing a comparative analysis across multiple dimensions.

Supervised Learning: Supervised learning algorithms learn from labeled training data to predict outcomes for unforeseen data [46]. The concepts of bias and variance are central to understanding the performance of these algorithms [47]. Bias refers to the error due to the model’s assumptions in the learning algorithm, while variance pertains to the error from the model’s sensitivity to fluctuations in the training set. An optimal balance between the two is crucial to avoid overfitting (high variance) and underfitting (high bias), both of which harm the model’s predictive accuracy on new data. This balance is often achieved through methods such as cross-validation and hyperparameter tuning. Ensemble methods, including bootstrapping and bagging, and boosting methods, aim to enhance model performance by combining several weak learners to form a stronger overall model [38].

Unsupervised Learning: Unlike supervised learning, unsupervised learning algorithms discern patterns in data without pre-existing labels, primarily through clustering and dimensionality reduction techniques [48]. Concepts such as data density and similarity measures are instrumental in these techniques, and specific algorithms, such as hierarchical clustering and DBSCAN, utilize these concepts [49]. Handling outliers, determining the appropriate number of clusters (e.g., via the elbow method or silhouette score), are other critical aspects of unsupervised learning [50].

Reinforcement Learning: In reinforcement learning, an agent learns to perform actions in an environment to maximize a reward through the process of trial and error [51]. This learning paradigm is defined in terms of a Markov decision process, where the agent’s actions in a certain state determine the probabilities of transitioning to other states [52]. The exploration versus exploitation dilemma is central to reinforcement learning, guiding the balance between trying new actions and exploiting known ones [53]. Techniques like Q-learning, SARSA, policy gradient methods, and more recent advances in deep reinforcement learning, provide frameworks to navigate this space [54].

Comparative Analysis: The three learning paradigms vary significantly in terms of data requirements, computational resources, interpretability, susceptibility to noise and outliers, and the level of human supervision required. For instance, supervised learning often requires large labeled datasets, which can be resource-intensive to collect, while unsupervised and reinforcement learning can work with unlabeled data. Supervised models are generally more interpretable than unsupervised and reinforcement learning models, making them suitable for applications where interpretability is a concern [55]. However, each of these learning paradigms has its strengths and weaknesses, and their appropriate usage largely depends on the specific problem at hand.

Understanding the subtleties of these paradigms is crucial for their effective application in environmental data analysis, given the complexity and high dimensionality of the data typically involved in this field [56].

4. Machine learning for enhanced IEQ assessment

In this section, we explore the practical applications of machine learning in assessing and enhancing Indoor Environmental Quality (IEQ) across a range of settings—residential, commercial, and specialized environments. We will examine how these algorithms contribute to optimized living conditions in homes, facilitate real-time environmental adjustments in commercial buildings, and ensure stringent IEQ compliance in specialized settings like healthcare facilities and data centers. Through these case studies, the section aims to provide an empirical substantiation of the machine learning methodologies discussed earlier, demonstrating their transformative potential in diverse IEQ contexts.

4.1 Residential buildings: Improving IEQ with machine learning

Machine learning has become a transformative force in improving Indoor Environmental Quality (IEQ) in residential buildings, offering significant advancements in Indoor Air Quality (IAQ), energy management, and occupancy detection.

Machine learning algorithms such as stochastic gradient boosting and support vector machines have demonstrated effectiveness in the domain of IAQ. These algorithms play a pivotal role in offering accurate predictions concerning room occupancy based on air quality indicators [57, 58]. Moreover, they are capable of efficiently classifying factors affecting IAQ under various conditions [59]. Other innovative approaches, including decision trees and hidden Markov models, have been applied to account for both real-time and anticipated occupancy states. These models have shown the capability to rectify sensor errors, thereby ensuring the reliability of IAQ parameters [60, 61].

Focusing on the challenges of sensor failures in pollutant monitoring, a significant study evaluated three Machine Learning (ML) algorithms – Multi-layer Perceptron (MLP), K-Nearest Neighbor (KNN), and Random Forest (RF) – for their effectiveness in classifying sensor readings as faulty or not. These algorithms demonstrated their superiority over standard statistical methods by creating better separation boundaries and utilizing contextual information. Numeric results from a 20-fold cross-validation displayed high average Area Under Curve (AUC) scores for each pollutant: 0.96 for MLP, 0.97 for KNN, and 0.97 for RF, indicating their robust performance in detecting faulty sensor readings [62].

Another key aspect of machine learning in IEQ is its application in energy efficiency. Studies have validated sensor data using techniques like artificial neural networks (ANNs) and Bayesian Networks, particularly focusing on U.S. commercial and residential buildings. Analyzing data from various sensors, these studies found that the root-mean-square error (RMSE) was generally less than 10% of the sensor’s mean value for most sensors. However, the energy sensor data showed higher RMSEs, often exceeding 200%. Bayesian Networks, despite requiring longer training times, yielded lower errors compared to ANNs. Comparative RMSEs for ANNs and Bayesian Networks were 17.7% vs. 13.5% for liquid pressure, 5.0% vs. 2.1% for humidity, 1.9% vs. 1.1% for temperature, 0.03 vs. 0.02 for water flow, and 0.89 vs. 0.85 for energy consumption, highlighting the potential of these methods in enhancing sensor accuracy and energy efficiency [63].

In the field of residential energy management, machine learning, particularly Random Forest algorithms, has shown impressive efficacy in predicting home occupancy based on thermostat data. This was evident in a study evaluating various machine learning models, including heuristic baselines, traditional classifiers, and sequential models like recurrent neural networks (RNNs), for their ability to predict home occupancy. The Random Forest algorithm was noted for its high accuracy across different time horizons and efficient training for individual edge devices. Key numeric results included training times of 0.38 seconds for the Random Forest and 10 seconds for the RNN, with inference times of 0.001 seconds and 0.01 seconds, respectively, underscoring the potential of machine learning in adaptive thermal control in residential buildings [64, 65].

Further advancements in residential IEQ have been made using ensemble techniques and ANNs for the precise prediction of heating and cooling loads, especially in air-conditioned residential buildings in warm climates. One particular study focused on reducing energy consumption through energy-efficient building design, employing machine learning methods to estimate heating and cooling loads based on physical building characteristics. The performance of these methods was assessed using metrics like Mean Absolute Error (MAE) and Root Mean Square Error (RMSE), with significant improvements noted in cooling load prediction. The lowest MAE was 0.390, and the highest R² value was 0.996, indicating the model’s high accuracy and efficiency [66, 67].

The integration of machine learning with smart home technologies has opened new avenues in IEQ management. For example, machine learning algorithms, used alongside IoT sensors, have enabled real-time monitoring and predictive analysis of indoor air pollutants and noise levels. Techniques such as fuzzy logic have been utilized to assess the efficacy of HVAC systems. These multifaceted applications of machine learning provide a robust foundation for enhancing IEQ in residential buildings and pave the way for a holistic approach to residential IEQ, marking a strong.

4.2 Commercial buildings: Real-time IEQ control with machine learning

Managing Indoor Environmental Quality (IEQ) in commercial buildings presents unique challenges due to varied occupancy patterns and the multifunctional nature of these spaces. The deployment of machine learning techniques has emerged as a key solution, offering real-time adaptability and predictive power for IEQ management.

Machine learning’s versatility in real-time monitoring, classification, and predictive modeling is evident from studies that utilized algorithms like decision tree, random forest, and Support Vector Machine (SVM) for air quality and noise level predictions. Notably, SVM showed significant effectiveness in air quality prediction, with its accuracy improving from 75–95% after data quality adjustments. In contrast, decision tree and random forest algorithms achieved accuracies of 76% and 82%, respectively. For noise data classification, SVM achieved a remarkable 98% accuracy, consistently outperforming other models over a six-year evaluation period. The critical role of data quality and normalization in improving prediction accuracy was highlighted, with CO and SO2 sensors achieving high accuracies, while PM2.5 sensor accuracy was comparatively lower [68].

Despite these advancements, gaps in the literature are evident. Some studies have developed forecast models for CO2 buildup without elaborating on real-time IEQ monitoring [69]. Others have explored innovative areas like integrating machine learning with EEG signals for predicting IEQ conditions based on occupants’ brainwave patterns [70]. Additional studies employed decision trees, hidden Markov models, and stochastic gradient boosting algorithms for immediate and future state occupancy prediction, demonstrating machine learning’s potential for real-time IEQ control [57, 60]. The advantages of machine learning over traditional statistical approaches were also noted in IAQ sensor reading classification [62].

Specialized applications in commercial settings have seen the use of neural network modeling for identifying contaminant source positions and Sparse Spectrum Gaussian Process Regression (GPR) for real-time air quality prediction. These applications highlight machine learning’s adaptability to various commercial environmental challenges [71, 72]. In educational and enterprise environments, studies have focused on ventilation rate control using IoT protocols and employing machine learning for air quality forecasting, further underscoring the diverse applicability of these technologies [73, 74].

Empirical evaluations often favor neural network-based models, using metrics such as RMSE, R2 score, and error rate. The ongoing advancements in these techniques, particularly in algorithm development and sensor calibration, promise enhanced precision and efficiency in monitoring indoor pollutants [75].

While machine learning offers promising opportunities for real-time IEQ control in commercial buildings, further research is needed for its effective implementation. Future research directions may explore the integration of machine learning with biometric signals for personalized environmental control [76].

4.3 Specialized environments: Strict IEQ regulation through machine learning

The challenge of maintaining optimal Indoor Environmental Quality (IEQ) in specialized environments, notably hospitals, has been rigorously researched. In such contexts, regulating IEQ is critical, impacting patient health, recovery, and satisfaction. Recognizing the limitations of conventional methods, researchers are increasingly turning to machine learning algorithms for more advanced computational approaches to these complexities.

Machine learning’s application in hospitals has been pivotal in enhancing patient satisfaction, particularly concerning IEQ. A significant study at King Abdullah University Hospital (KAUH) in Jordan utilized Support Vector Machines (SVM) with a linear kernel and K-Nearest Neighbors (K-NN) to predict patient satisfaction. The SVM model, with a configuration of C = 0.01, proved more effective than other SVM kernels and K-NN. This effectiveness is evidenced by SVM’s Pearson R-square value of 0.8948 with a P-value of .0001, surpassing K-NN’s Pearson R-square of 0.2984. The study involved gathering self-reported data and field monitoring of environmental indicators within patient rooms at KAUH, using the same dataset for both training and testing. These outcomes highlight machine learning’s role in providing insights that impact healthcare outcomes, particularly in enhancing IEQ to improve patient well-being [77].

Data collection in these environments often involves a mix of methodologies, including self-reported data, field monitoring, and sensor networks, creating diverse datasets for machine learning training and validation. For example, Elnaklah et al. combined sensor-generated data and human-reported satisfaction metrics to amass a comprehensive dataset [78].

Several machine learning models have shown effectiveness in regulating IEQ in hospital settings. An Autoregressive Hidden Markov Model (ARHMM) developed for a laboratory equipped with a sensor network demonstrated an average estimation accuracy of 80.78% in predicting occupancy patterns. This model, utilizing data from passive infra-red (PIR) sensors, CO2 concentration sensors, and relative humidity (RH) sensors, was benchmarked against classical Hidden Markov Models (HMM) and Support Vector Machines (SVM).

Bayesian networks have also been used to assess the risk of symptomatological complaints related to poor IEQ in Intensive Care Units (ICUs). A study in nine ICUs in João Pessoa, Brazil, analyzed temperature, noise, lighting, and air quality data, along with professional interviews. The Bayesian model revealed a 42.2% probability of physical symptoms and a 45.3% probability of psychological symptoms from environmental discomfort among ICU employees, with environmental temperature identified as a significant impacting factor [79, 80]. These probabilistic graphical models provide insightful perspectives on potential risk areas in IEQ within hospital settings.

Notably, there are gaps in existing literature, with several studies investigating IEQ through questionnaires and field studies without employing machine learning. For example, a study examining apartment and office buildings did not leverage machine learning, representing a missed opportunity for data-driven insights [81]. This highlights the significant potential for integrating machine learning techniques in future studies to provide more accurate, actionable recommendations for IEQ regulation in hospitals.

5. Conclusions

The chapter has provided an in-depth exploration of the integration of Machine Learning (ML) techniques into Indoor Environmental Quality (IEQ) assessment. IEQ, an aspect critical to human well-being, is intricately linked with numerous parameters including air quality, thermal, acoustical, and visual comfort. The chapter delineated the limitations of traditional methods in IEQ assessment, especially the lack of real-time data and predictive capabilities, and suggested machine learning as a potent solution for augmenting both the scope and accuracy of such evaluations. The implications are far-reaching, impacting residential buildings, commercial structures, and specialized environments such as healthcare facilities.

While machine learning appears to offer a robust approach to managing and enhancing IEQ, there are questions that remain unanswered. For instance, the ethical considerations of data collection and usage, particularly in specialized environments like healthcare facilities, warrant attention. Similarly, the interpretability of complex machine learning models and the implications for regulatory compliance are subjects requiring more in-depth study.

5.1 Future research directions

Data Ethics and Privacy: As machine learning increasingly intersects with IEQ assessment, ethical considerations around data collection and use will become more pressing. Future studies could focus on the ethical guidelines that ought to govern this intersection.

Interpretable Machine Learning Models: Future work could concentrate on developing more interpretable models without sacrificing predictive power, which is crucial for gaining regulatory approval and social acceptance.

Adaptive Systems: Research could delve into the development of adaptive machine learning systems that can evolve in real-time to meet the dynamic nature of IEQ elements, thereby offering more precise control mechanisms.

Personalization: Given the inter-individual differences in comfort and health responses, studies focusing on personalized IEQ assessment and control algorithms could offer a more nuanced approach to managing indoor environments.

Longitudinal Studies: To robustly assess the impact of machine learning on IEQ over time, future research could employ longitudinal study designs, perhaps integrated with natural experiments in real-world settings.

Cross-Sector Collaboration: As machine learning and IEQ assessment are topics with multidisciplinary ramifications, partnerships between AI researchers, environmental scientists, healthcare professionals, and policy-makers are essential for holistic solutions.

Resource Optimization: Machine learning algorithms capable of balancing both IEQ and energy efficiency would be a significant advancement, addressing the practical constraints of implementing ML in building management systems.

Global Standards: The development and adoption of global standards for machine learning in IEQ assessment could further facilitate interoperability and effectiveness across diverse building types and geographical locations.