Prediction of video game age labels with multimodal biosignals and artificial intelligence

Today, video games have managed to become the most preferred entertainment tool. Therefore, the number of video game players has reached almost half of the world's population. Having a number of player populations all over the world, video games have the potential to affect society positively or negatively. Video games can have different impacts on players ranging from different age groups. Hence, institutions or companies rate the video game to label it for the suitability of the game content for the age of players. There are no companies, institutions or systems to label the video games in Turkey. Video game age labeling is usually carried out employing methods such as questionnaires, interviews and observations, which are highly subject to subjectivity and bias. Recently, it is seen that alternative or additional objective methods are used to these approaches. Generally, biosignal measurement methods are used as an objective method. In the research, it is aimed to design a new age labeling system in which video game age labeling can be performed by using multimodal biosignals and artificial intelligence methods. For this purpose, the most popular action genre video game trailers with PEGI age labels were shown to the participants. During the process of watching the video game trailer by participants; brain signals with Electroencephalogram (EEG), heart rate and blood oxygen amount with Oximeter, skin response with Galvanic Skin Sensor (GSR), and signals of body movements with Inertia Measurement Unit (IMU) were simultaneously collected and time-stamped. The obtained raw data were labeled with the video game age labels. The labeled data were resampled by utilizing interpolation and upsampling methods. Then, the data were combined using the sliding window technique. The combined data were respectively preprocessed, divided into training/test, and their features were selected using different feature selection algorithms. Afterwards, artificial intelligence models were established with Adaboost, kNN, NB, RF, MLP and SVM algorithms to estimate the age labels. After running and testing the models, the best prediction performance was obtained using the RF model with 96.6%. As a result of the research, a system that is the first in the world has been developed in which age labels of video games are labeled in real time with multi-modal biosignals and artificial intelligence. When this developed system is compared with manual age labeling methods, it has been observed that the performance of age labeling with biosignals and artificial intelligence is higher. Furthermore, in this study, it has been determined that the best distinguishing features of video game age labels are respectively MOT.AccZ signal which gives the movement in the Z-axis of the head obtained from the IMU source, POW.O2.Theta signal obtained from the right occipital lobe source of the brain, and POW.F7.Gamma signal obtained from the frontal lobe source of the brain.