Learner’s emotion prediction using production rules classification algorithm through brain computer interface tool
Enhancements in cognitive neuroscience and brain imaging technologies such as Human-Computer Interaction (HCI) have started to provide human with the ability to interact directly with the brain. The use of sensors known as Brain-Computer Interface (BCI) tool can monitor the physical processes and me...
| Summary: | Enhancements in cognitive neuroscience and brain imaging technologies such as Human-Computer Interaction (HCI) have started to provide human with the ability to interact directly with the brain. The use of sensors known as Brain-Computer Interface (BCI) tool can monitor the physical processes and mental states that occur in the human brain. Emotion recognition is found to be an important aspect of the interaction between human being where emotion influences human in daily life. Researchers investigated many methods to capture and recognise emotion, such as through speech, facial expression, and physiological signals. Electroencephalogram (EEG) signals are found to be the best physiological signals that contain valuable information about human mental state. However, these EEG signals involve a lot of data and need to be mined efficiently in order to make it valuable and meaningful. The crucial parameter of emotion recognition has largely been ignored because it is always misunderstood and is hard to measure. No EEG studies in Malaysia has been done on school children to study their emotional behaviour while learning. Classification and prediction are the functions provided by the data mining techniques that suit in EEG signal processing. The objectives of this research are to classify the user emotion characteristics by using EEG signals based on children’s behaviour, to develop a prototype of an emotion prediction system named as MYEmotion and to validate the developed prototype in predicting the positive and negative emotions of the children. 16 datasets of attention and meditation levels were collected from a qualitative sampling of 10 years old school children in Pekan, Pahang using a BCI headset tool. Each respondent underwent two mathematical game sessions using a smartphone with a two-minute break in between each session. From the data analysis using WEKA software, the production rules classifier (PART) is found to be the most accurate classification algorithm in classifying the emotion which yields the highest precision percentage of 99.6% compared to J48 (99.5%) and Naïve Bayes (96.2%). The decision lists generated by PART classifier that represent the regularities of the attention and meditation levels among children are converted into several rule sets of positive and negative emotions. These rule sets was implemented in the MYEmotion using MATLAB environment. MYEmotion summarises the entire procedure starting from the pre-processing to the end. A baseline set which is adopted from an established eSense meter value was also coded into the prototype. The data analysis of the both baseline and rule-based prediction sets have shown that there are not many differences between the trend of the positive and negative emotions percentage of both sets. The reliable relationship between EEG signals of the attention and meditation and their impact towards the positive and negative emotions among children while learning illustrates the potentials in detecting mental states which are relevant to tutoring such as comprehension, engagement and learning impact. In future, this research can be an initial work in automating tutorial decisions in an intelligent tutoring system which are able to adapt to the behaviour of the learners based on the detected mental states. Therefore, more relevant information about the students can be provided to the schools and teachers in order to increase the learning impacts. |
|---|