Benchmarking Machine Learning Techniques for Attention Detection

Being attentive among the chaos of the current world is very important. It has applications in various fields such as education, safety, healthcare. Attention patterns can be very helpful in diagnosing mental health conditions such as ADHD, depression and can lead to early treatment. It is also very important in workplaces which involve high-risk tasks that need the full attention of the person who perform it to prevent any mishaps and prevent any damages. It is also important in driver monitoring systems and can play a pivotal role in preventing accidents and dangerous situations. This paper compares various machine learning models which include KNN, Naive Bayes, SVM and 2 tree based models which are RandomForest and XGBoost, by observing how accurately they can detect attentive and inattentive states. Each model has been trained on an extensive dataset containing over 4000 records and have been cross-validated for the best possible accuracy and fair comparison between models. The dataset contains various features obtained from video-feed that have undergone extensive pre-processing to obtain numerical values. The final dataset contains over 4000 records of data such as coordinates and confidence values in numerical format and categorical variables such as pose and label. The results of the comparison have been displayed using numerical statistics such as accuracy, precision, recall and also using graphical means such as ROC curves and confusion matrices.