IoT-enabled ambient monitoring: intelligent indoor air quality assessment and prediction

Indoor Air Quality (IAQ) has emerged as a critical factor influencing public health,productivity, and overall well-being, especially as urban populations spend approximately80-90% of their time indoors. Poor Indoor Air Quality Index (IAQI) has been directlylinked to a wide range of health issues, including respiratory conditions, fatigue, and cognitiveimpairments, and its impact is amplified in vulnerable populations such as children,the elderly, and individuals with pre-existing conditions like asthma. In recent years, advancementsin technology have enabled the development of Internet of Things (IoT)-basedmonitoring systems capable of efficiently detecting and reporting environmental conditions,and even predicting future air quality trends. However, most of these systems focus solelyon processing specific environmental parameters, lacking a comprehensive framework thatevaluates how air quality directly impacts human conditions, such as thermal comfort.This thesis proposes and validates a novel, multi-stage methodological framework thattransforms IAQ assessment from simple environmental monitoring into a predictive, occupantcentric practice. The feasibility and performance of each stage of the framework aredemonstrated, showcasing a logical progression from foundational monitoring methods toadvanced algorithmic applications for health.The research progresses through three interconnected phases. First, we developed anIoT-enabled IAQ monitoring system capable of real-time Carbon dioxide (CO2) monitoringand prediction. This system utilizes the Message Queuing Telemetry Transport (MQTT)protocol for efficient real-time data transmission and incorporates machine learning modelsto predict CO2 steady-state levels. The core novelty lies in combining a mathematical massbalancemodel with Long Short-Term Memory (LSTM) networks to forecast indoor CO2steady-state concentrations, achieving an error margin of only 5.5%.Second, the research introduces an innovative update to existing IAQ indexing to makeit more meaningful. Building on Marino’s time-fraction methodology, the primary noveltywas to replace subjective assumptions with a data-driven approach, using a machine learningmodel to update and assign the weighting parameters for key environmental factors.A supporting model was first developed to predict the Predicted Mean Vote (PMV) usingonly temperature and humidity, achieving 74.2% accuracy within the ±0.5 thermal comfortrange. We then used a Random Forest model to predict suitable air parameter weightsand applied these to the time-fraction-based IAQI model based on Marino. It showed astronger correlation with occupant comfort (R =0.65) than a standard commercial sensor’sindex (R =-0.13), demonstrating a validated method for creating more human-centric IAQassessments. Furthermore, when validated against a simplified, equally-weighted baseline,the proposed index also proved more effective at identifying days with poor air quality,achieving a higher Average Precision score (0.549 vs. 0.496).The third stage presents a novel algorithmic contribution for health risk forecasting.A Spatio-Temporal Graph Convolutional Network (ST-GCN) was developed to effectivelyfuse heterogeneous data types, including environmental factors, biosignals, and static patientprofiles. Validated on a public asthma dataset, this model achieved 91.9% accuracyin predicting daily symptoms and a robust 86% accuracy without inhaler data, showcasingthe algorithm’s potential for health prediction tasks.Collectively, this thesis establishes and validates a comprehensive methodological pathwayfor developing intelligent indoor environmental systems. Its core novelty lies in demonstratinga complete, data-driven progression from proactive environmental forecasting, tothe creation of a validated, human-centric assessment index, and culminating in the developmentof a novel algorithm that validates how the principles of multi-source data fusioncan be applied to achieve high-accuracy, personalized health risk prediction. This work providesa verifiable blueprint for transforming IAQ monitoring from a simple data collectionexercise into an actionable, occupant-focused practice.