Predicting Anger Proneness Using Deep Learning Techniques

High levels of anger proneness can affect the individuals and others around them in the society. Anger proneness or the likelihood of experiencing anger in response to certain stimuli or situations is influenced by a variety of factors, including personality traits, life experiences, and environmental factors. Accurately predicting the level of anger proneness in individuals can help mitigate a variety of risks, including relationship and mental health problems. In this paper, we tried to create a model that is capable of accurately predicting levels of anger proneness based on one’s answers to a personality questionnaire. To do so, we utilized a deep learning algorithm namely SAINT (Self-Attention and Intersample Attention Transformer), alongside with other three machine learning and deep learning algorithms as the comparisons, LSTM (Long Short-Term Memory), GRU (Gated Recurrent Unit), and XGBoost (EXtreme Gradient Boosting). The four models were trained with the same dataset and the same data preprocessing methods. All the four algorithms couldn’t manage to produce any models with satisfactory performances, with the highest performing model produced using the state-of-the-art machine learning classifier XGBoost only achieving 55.38% accuracy. Multiple data preprocessing methods such as resampling and feature selection with the help of XGBoost were applied to the dataset with negligible effects to the models’ performances. The results lead us to the conclusion that the dataset contains too many inconsistencies in its rows, as can be fairly expected from a dataset based on individuals’ responses to an online personality questionnaire. The inconsistencies affected all the models’ ability to find the patterns in between rows and columns and compromise their resulting efficacies to do the inferences.