Prediction of Wi-Fi Comfort in Buildings using Support Vector Machines

Comfort of occupants in semi-public buildings depends on ambient parameters such as temperature, relative humidity, carbon dioxide concentration, light intensity, noise, cleanliness, etc. In addition to the above parameters, occupants' indoor comfort is observed to be significantly influenced by the quality of Wi-Fi connection provided for ubiquitous internet access in buildings. Hence, in this work, another important dimension to indoor comfort is introduced-Wi-Fi comfort. A novel comfort management architecture is proposed which employs machine learning to recognize the level of Wi-Fi comfort. The proposed architecture complements the extensive research done in the area of network planning, by providing Quality of Experience (QoE) of Wi-Fi as an effective input to such mechanisms. Comfort is predicted based on the Wi-Fi signal strength, Wi-Fi access permissions, steady-state occupancy and applications used by the occupants. A case study is presented in which occupants use a specific set of applications, such as email, voice and/or video for a given duration. The Support Vector Machines (SVM) method is used for classification of comfort levels. The performance evaluation of the classifier and of the overall system is discussed in detail.