VSIT Student Behaviour Classification Dataset

The study utilizes a student behavioural dataset comprising 3,000 student records collected from Vidyalankar School of Information Technology (VSIT), Mumbai, India, under the 2024-2025 academic year to investigate the automatic classification of student behavioural states using textual behavioural descriptions and structured academic, demographic, and behavioural characteristics. Data were collected from the various academic communication tools in the institution such as Learning Management System (LMS), discussion forums, the reflection on assignments, classroom interaction, and students' feedback. To maintain privacy and confidentiality all personally identifiable information was removed prior to the analysis. Doccano-based manual annotation was performed using predefined behavioural guidelines to consistently assign each behavioural text instance to one of the four target categories: Engagement, Confusion, Stress, or Neutral. The dataset contains 18 attributes, including one student identifier, 15 input features, one behavioural text field, and one target class. The structured attributes consist of demographic, academic, and behavioural-profile features as shown in Table 3. The distribution of the student behavioural dataset across the four behavioural classes and the corresponding training, validation, and independent testing subsets is presented in Table 3. Table 3. Dataset characteristics and distribution Category Features Type Role Identifier Student_ID Identifier Excluded from modelling Demographic Age Numerical Input Demographic Gender Categorical Input Academic Program Categorical Input Academic Semester Categorical Input Academic Attendance Numerical Input Academic CGPA Numerical Input Academic Internal_Marks_25 Numerical Input Academic Assignment_10 Numerical Input Academic Quiz_10 Numerical Input Academic Midterm_Marks_20 Numerical Input Behavioural Motivation Numerical Input Behavioural Self_Awareness Numerical Input Behavioural Stress_Management Numerical Input Behavioural Social_Skills Numerical Input Behavioural Time_Management Numerical Input Textual Behaviour_Text Text Input Target Behaviour_Class Categorical Ground truth Dataset distribution Behaviour class Original Testing Training Validation Engagement 156 31 100 25 Confusion 279 55 179 45 Stress 1,158 232 741 185 Neutral 1,407 282 900 225 Total 3,000 600 1,920 480