Machine Learning-Based Analysis of Digital Ghosting and Deceptive Behaviour via Keystroke Dynamics

Digital communication has made interpersonal interaction fast and accessible, but it has also made certain behavioural changes difficult to interpret.Digital ghosting refers to a sustained withdrawal from communication without a clear explanation, whereas deceptive behaviour involves intentionally providing false or misleading information.This paper proposes a machine-learning research framework that examines whether keystroke dynamics and communicationpattern features can support the analysis of these behaviours.Keystroke features include dwell time, flight time, typing speed, pauses, error rate, backspace frequency, and response time.Communication features include response delay, interaction frequency, unanswered messages, and deviations from an individual's usual pattern.The framework considers separate prediction tasks for potentially deceptive responses and possible communication withdrawal.Candidate classifiers include Logistic Regression, Decision Tree, Random Forest, Support Vector Machine, Gradient Boosting, and Neural Networks where the amount of data permits.Evaluation is planned using accuracy, precision, recall, F1-score, confusion matrices, and ROC-AUC, with participant-level data separation to reduce leakage.As this is a proposed study, no experimental performance results are claimed.The paper emphasizes that typing changes are not proof of deception or ghosting and that any output must be interpreted as an uncertain behavioural indication.Consent, anonymization, data minimization, explainability, and secure storage are central design requirements.