CO2-based recommendation system for predictive ventilation schemes in different real-life building applications using deep learning

Traditional building management systems often rely on extensive sensor data, which can be impractical and costly due to limited data availability. This study addresses the challenge of effectively managing indoor conditions with minimal infrastructure by developing a data-driven, CO 2 -based ventilation recommendation system. A minimally invasive predictive model is developed, using historical CO 2 measurements to forecast future concentrations and provide timely ventilation recommendations. By focusing solely on CO 2 data, this approach reduces the need for additional IoT sensors, lowering implementation costs while maintaining optimal indoor conditions. The performance of various machine learning and deep learning models is systematically evaluated, demonstrating the efficiency of this simplified solution for real-time building environment management. The best-performing model achieved a Root Mean Square Error of 50.46 ppm and an R 2 of 0.97 for short-term predictions in large rooms, while maintaining robust accuracy across different environments. The methodology, evaluated across diverse building types, demonstrates that focusing solely on CO 2 data can provide accurate ventilation recommendations, making the system scalable and applicable to real-world scenarios.