Machine Learning–Based Prediction of Occupant Symptom Risk Using Indoor and Outdoor Environmental Quality Data in University Dormitories

Indoor air quality (IAQ) is a major determinant of occupant health, with people spending over 90% of their time indoors. This study presents a machine learning–based classification framework for predicting self-reported health symptoms in university dormitory residents, using continuous indoor and outdoor environmental monitoring data and occupant survey responses collected across 7 monitoring periods from May 2024 to June 2025 in 2 student accommodation facilities in Milan, Italy. From 74 occupant samples, 131 candidate predictors were derived from 5 indoor and 5 outdoor environmental parameters together with occupant-level and contextual variables and reduced to symptom-specific feature sets of 3–40 predictors through permutation importance. Seven symptoms each reported by at least 10 occupants (sneezing, headache, cough, dry skin, fatigue, dry/sore throat and runny nose) were modelled as independent binary classification tasks under leave-one-period-out cross-validation (LOPO-CV), with the best classifier selected from 23 candidates per symptom based on balanced accuracy and thresholds set individually by Youden′s J index. Mean AUC across the 7 models was 0.595 (range: 0.398–0.705), with 4 symptoms exceeding AUC 0.6; dominant predictive signals were drawn from outdoor environmental statistics and distributional shape features rather than mean indoor pollutant concentrations. At a combined accuracy criterion of ≥ 5/7 symptoms correctly classified, 58.1% of occupants were correctly profiled across all 74 samples, rising to 64.3% in the symptomatic subgroup (n = 42); excluding the 2 symptoms with the lowest model specificity raised ≥ 3/5 combined accuracy to 92.0% across the full sample and 90.0% in the symptomatic subgroup. The results demonstrate the feasibility of multisymptom environmental risk profiling and support the concept of a learning building, in which continuously updated occupant health models can inform proactive and personalised indoor environment management.