Predictive analysis of college students' academic procrastination behavior based on a decision tree model

The study investigates how college students’ academic procrastination was affected during the COVID-19 crisis and aims to identify key predictors to inform interventions that improve academic performance and well-being. Prior work highlights that psychological well-being, life satisfaction, negative emotions, self-esteem, life autonomy, smartphone addiction, and environmental and contextual factors (e.g., sense of school belonging, academic achievement) relate to procrastination. The authors note a gap in predictive modeling of academic procrastination during crises and propose using an interpretable decision tree (C5.0) to predict high vs. low procrastination and rank predictor importance. The research question centers on which biopsychosocial factors best predict academic procrastination among Chinese college students during the pandemic and how accurately a decision-tree model can classify students’ procrastination levels.

The review defines academic procrastination as intentional delay of academic tasks despite negative consequences, encompassing behavioral, cognitive, and emotional dimensions. Drawing on the Biopsychosocial Model, eight hypothesized predictors are outlined: subjective well-being (SWB), smartphone addiction, negative emotions (depression, anxiety, stress), self-esteem, life autonomy, pro-environmental behavior, academic achievement, and sense of school belonging. Evidence suggests SWB improves self-regulation and resilience, predicting lower procrastination (H1: SWB negatively predicts procrastination). Smartphone addiction, linked to poor self-regulation, attention deficits, and sleep problems, predicts higher procrastination (H2: positive prediction). Negative emotions impair cognition and goal-directed behavior, increasing procrastination (H3: positive prediction). Self-esteem supports motivation and emotion regulation, predicting lower procrastination (H4: negative prediction). Life autonomy promotes intrinsic motivation and self-management, predicting lower procrastination (H5: negative prediction). Pro-environmental behavior relates to planning, executive function, and self-regulation, predicting lower procrastination (H6: negative prediction). Academic achievement correlates with self-efficacy and lower procrastination (H7: negative prediction). Sense of school belonging supports engagement and reduces procrastination, particularly threatened during online learning and lockdowns (H8: negative prediction). The decision tree approach is justified for its interpretability and ability to capture interactions among predictors.

Design and setting: Cross-sectional predictive study using a decision tree (C5.0). The study was conducted in the Guangxi Zhuang Autonomous Region, China, noted for proactive academic policies during COVID-19. Data were collected from September 7 to October 15, 2022. Sampling and participants: Three-stage random sampling. First, 3 of 83 regional universities were randomly selected; only one granted authorization due to lockdown restrictions. Freshmen had not started, and many seniors were off-campus; thus, sophomores were targeted across 46 majors (~15,000 students). Using an online random number generator, 800 sophomores were invited; 24 invalid responses were removed, yielding 776 valid cases. Demographics: 219 males (28.2%), 557 females (71.8%); ages 19–25. Procedures: Due to campus access restrictions, a trained university counselor assisted. After informed consent and study briefing (confidentiality, right to withdraw), students scanned a QR code to complete an online survey (~20 minutes). Assistance was available for clarification. Measures (5-point Likert scales unless noted; higher scores indicate higher construct levels unless specified): - Subjective Well-Being (SWB; Diener; Chinese version Xing, 2002), 20 items; α = 0.860. - Smartphone Addiction Scale—Short Version (SAS-SV; Kwon et al., 2013), 10 items; α = 0.840. - Depression, Anxiety, Stress Scale (DASS-21; Lovibond & Lovibond; Antony et al., 1998), 21 items; α = 0.965. - Rosenberg Self-Esteem Scale (RSE; Rosenberg, 1965), 10 items; α = 0.714. - Life Autonomy subscale (Pan & Xie, 2010), 12 items; items 7–12 reverse-scored; higher raw scores reflect lower autonomy; α = 0.946. - Pro-Environmental Behavior (Liu & Wu, 2013), 11 items (public/private domains); α = 0.953. - Psychological Sense of School Membership (PSSM; Goodenow; Chinese version Pan et al., 2011), 18 items; α = 0.838. - Academic Procrastination (Tuckman, 1991), adapted to 5-point scale, 16 items; α = 0.920. Demographic variables: gender, age, academic achievement. Descriptive statistics: Means (SD) reported. Academic procrastination M = 2.593 (SD = 0.635), below the 3.0 cutoff (60% of full score). Data coding: All variables dichotomized at 60% of full score (score >3 coded 1; ≤3 coded 0). Distribution examples (N=776): academic procrastination low 81.06%, high 18.94%; SWB high 77.06%; smartphone addiction high 23.84%; negative emotion high 11.86%; self-esteem high 63.66%; life autonomy high 72.04%; pro-environmental behavior high 67.53%; academic achievement high 38.40%; school belonging high 56.06%. Modeling approach: IBM SPSS 26.0 for descriptive analyses; IBM SPSS Modeler 18.0 for decision tree modeling using C5.0 (extension of ID3/C4.5). The C5.0 algorithm selects splits by maximum Gain Ratio based on entropy; post-pruning applied to reduce error. Data split into 70% training (n=544) and 30% testing (n=232) per Gholamy et al. (2018). Performance metrics included accuracy, precision, and recall. The target variable was binary academic procrastination (high vs. low).

- Eight predictors of academic procrastination were identified and ranked by importance: 1) subjective well-being (most important), 2) smartphone addiction, 3) negative emotions, 4) self-esteem, 5) life autonomy, 6) pro-environmental behavior, 7) academic achievement, 8) sense of school belonging (least important). Figure 3 reports relative importance values for some variables (e.g., self-esteem 0.081; life autonomy 0.059; pro-environmental behavior 0.058; academic achievement 0.010; school belonging 0.008), with SWB, smartphone addiction, and negative emotion highest. - Model performance: Training accuracy 87.50% (476/544 correct). Testing accuracy 85.78% (199/232 correct). - Confusion matrices: • Training: low actual predicted low 410, predicted high 25; high actual predicted low 43, predicted high 66. • Testing: low actual predicted low 182, predicted high 12; high actual predicted low 21, predicted high 17. - Precision and recall (testing set): • Low procrastination: precision 90.24%, recall 94.12%. • High procrastination: precision 69.17%, recall 56.46%. - Class distribution: Overall, low procrastination 81.06%, high 18.94%; in the training root node, low 79.963%. - Descriptives (means on 1–5 scale): SWB 3.380, smartphone addiction 2.724, negative emotion 2.122, self-esteem 3.430, life autonomy 3.543, pro-environmental behavior 3.478, academic achievement 3.070, sense of school belonging 3.257, academic procrastination 2.593.

The model indicates that psychological well-being and emotion/behavior regulation factors are central to academic procrastination during crises. Higher subjective well-being appears to foster conscientiousness, time monitoring, and adaptability, reducing procrastination. Smartphone addiction undermines attention and self-control, promotes avoidance coping, and can impair sleep, all of which increase procrastination—effects exacerbated by heavy smartphone dependence during online learning. Negative emotions (anxiety, stress, depression) impair executive functions and goal-directed behavior, elevating procrastination. Self-esteem enhances motivation, academic confidence, and emotion regulation, indirectly curbing procrastination. Life autonomy supports self-direction and self-control; neuroscience evidence links enhanced prefrontal control to reduced procrastination. Pro-environmental behavior reflects planning, self-regulation, and execution skills that oppose procrastination tendencies. Academic achievement and sense of school belonging showed relatively smaller importance; during crises, students may prioritize health and family over academics, and prolonged online learning may weaken school connectedness. Nonetheless, these factors remain relevant for overall academic engagement and mental health and may exert indirect effects via mediators such as self-efficacy and emotions.

Using a decision tree (C5.0), the study accurately predicted academic procrastination (testing accuracy 85.78%) and ranked predictors. Subjective well-being, smartphone addiction, and negative emotions emerged as core predictors, underscoring the pivotal roles of psychological well-being, technology use, and emotional health. Self-esteem, life autonomy, pro-environmental behavior, academic achievement, and school belonging also contributed to prediction and should inform comprehensive interventions. The findings provide actionable insights for reducing academic procrastination by promoting well-being, enhancing self-regulation, and supporting students’ connection to their institutions, with implications during crises and beyond.

The cross-sectional design limits causal inference and understanding of changes over time. All participants were from a single university in the Guangxi Zhuang Autonomous Region, limiting generalizability. Future research should employ longitudinal designs to observe dynamics among predictors and procrastination, recruit from multiple regions and institutions to improve external validity, and examine additional variables not included here to broaden understanding and inform interventions.