Wearable-based human flow experience recognition enhanced by transfer learning methods using emotion data.
Computers in biology and medicine November 1, 2023 Muhammad Tausif Irshad, Frédéric Li, Muhammad Adeel Nisar et al.
Using physiological data from 25 subjects wearing the Empatica E4 wristband, Emotiv Epoc X EEG headset, and Biosignalplux RespiBAN during arithmetic and reading tasks, feature engineering and deep feature learning approaches automatically discriminated between flow and non-flow states. EEG sensor modalities alone achieved 64.97% accuracy and a macro averaged F1 score of 64.95%; fusing all sensor modalities yielded 73.63% accuracy and an AF1 of 72.70%. A transfer learning approach using emotional arousal classification on the DEAP dataset improved performance to 75.10% accuracy and an AF1 of 74.92%. The results suggest effective discrimination is possible with multimodal sensor data, and the success of transfer learning indicates that emotions and flow are connected.