This paper comprehensively investigates the performance of various strategies for predicting CO2 levels in school classrooms over different time horizons by using data collected through IoT devices. We gathered Indoor Air Quality (IAQ) data from fifteen schools in Navarra, Spain between 10 January and 3 April 2022, with measurements taken at 10-min intervals. Three prediction strategies divided into seven models were trained on the data and compared using statistical tests. The study confirms that simple methodologies are effective for short-term predictions, while Machine Learning (ML)-based models perform better over longer prediction horizons. Furthermore, this study demonstrates the feasibility of using low-cost devices combined with ML models for forecasting, which can help to improve IAQ in sensitive environments such as schools.
Indoor CO2 forecasting supports proactive ventilation control that balances air quality with energy efficiency. While Machine Learning (ML) models have shown strong performance in controlled settings such as schools, their generalization across indoor spaces with diverse occupancy dynamics remains poorly characterized. We present a systematic benchmark of 11 forecasting models spanning simple baselines, statistical methods, classical ML, deep learning, ensembles, and foundation models using 18 weeks of IoT sensor data spanning six real-world use cases: conference rooms, dining halls, hospitals, food markets, offices and student residences. Performance depends strongly on the prediction horizon and on the regularity of occupancy-driven CO2 patterns. Simple baselines tend to perform best at short horizons (10 min ahead), while ensembles and fine-tuned foundation models provide more robust accuracy at longer horizons (4 h ahead). Remarkably, zero-shot foundation models demonstrate the ability to outperform trained classical models in data-scarce scenarios, challenging the traditional paradigm of localized training. These findings indicate that optimal forecasting strategies are context-dependent and challenge the assumption of universal model superiority.