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

Academic procrastination is a significant issue impacting college students' well-being and academic performance, particularly exacerbated by public crises like the COVID-19 pandemic. This study addresses the limited research on predicting academic procrastination during such crises. The research question is: What factors predict academic procrastination behavior among college students, and how can these be modeled using a decision tree algorithm? The study's purpose is to construct a predictive model using the decision tree algorithm to forecast academic procrastination, considering various relevant factors. The importance of this study lies in its potential to provide higher education institutions with essential academic support and decision-making strategies to promote student psychological health and improve academic outcomes. Understanding the predictive factors allows for the development of targeted interventions to address this prevalent issue.

The literature review examines the existing definitions and conceptualizations of academic procrastination. While there is no single, universally accepted definition, most scholars agree that it involves the intentional postponement of academic tasks despite awareness of negative consequences. The review also explores various dimensions of academic procrastination, including behavioral, cognitive, and emotional aspects. The study then reviews the literature on predictors of academic procrastination, focusing on the biopsychosocial model. The selected predictors include subjective well-being (negatively correlated), smartphone addiction (positively correlated), negative emotions (positively correlated), self-esteem (negatively correlated), life autonomy (negatively correlated), pro-environmental behavior (negatively correlated), academic achievement (negatively correlated), and sense of school belonging (negatively correlated). The review highlights the rationale for choosing these factors based on prior research and their relevance to the context of a public health crisis. Finally, it establishes the suitability of the decision tree model for predicting academic procrastination, emphasizing its interpretability and accuracy in previous studies.

This quantitative study employed a three-stage random sampling method to select 776 sophomore students from one university in Guangxi Zhuang Autonomous Region, China. Data were collected using an online questionnaire consisting of demographic information and several scales measuring the selected predictor variables and academic procrastination. The scales used included: Subjective Well-Being Scale (SWB), Smartphone Addiction Scale (SAS-SV), Depression, Anxiety, and Stress Scale (DASS-42), Rosenberg Self-Esteem Scale (RSE), Life Autonomy Scale, Pro-environmental Behavior Scale, and Psychological Sense of School Membership (PSSM) scale, and the Academic Procrastination Scale. The Cronbach's alpha coefficients were reported for each scale, indicating acceptable reliability. Data were analyzed using SPSS 26.0 for descriptive statistics and Modeler 18.0 for decision tree modeling using the C5.0 algorithm. The dataset was split into 70% for training and 30% for testing the model. Academic procrastination was categorized as high or low based on a 60% cutoff score. Information entropy and Gain Ratio were used in the decision tree construction, and post-pruning was employed to optimize the model. Model evaluation included accuracy, precision, and recall rates.

The decision tree model achieved an accuracy of 87.50% on training data and 85.78% on testing data, indicating a relatively high predictive power. The model identified eight predictive factors of academic procrastination in order of importance: subjective well-being, smartphone addiction, negative emotions, self-esteem, life autonomy, pro-environmental behavior, academic achievement, and sense of school belonging. For the test data, the model's precision rate for predicting low academic procrastination was 90.24%, with a recall rate of 94.12%. The mean value of academic procrastination was 2.593 (SD = 0.635), indicating that less than half of the students exhibited high levels of procrastination. Table 1 presents the coding and descriptive statistics for each variable. The model visualization (Figure 2) illustrates the decision-making process and the relative importance of each predictor. Figure 3 provides a bar graph showing the importance of each predictive variable in the model.

The findings support the significant predictive role of subjective well-being, smartphone addiction, and negative emotions on academic procrastination. Students with higher subjective well-being tend to exhibit better self-regulation and time management, reducing procrastination. Conversely, smartphone addiction and negative emotions are associated with decreased self-control and attention, leading to increased procrastination. The less significant roles of self-esteem, life autonomy, pro-environmental behavior, academic achievement, and sense of school belonging suggest that while important, these factors may be less impactful during times of crisis when other concerns (health, financial) take precedence. The study's findings align with previous research on the correlation between these variables and procrastination. The relatively lower importance of academic achievement and sense of school belonging during the pandemic might be attributed to the disruption caused by online learning and other pandemic-related issues that impacted students' priorities. The model's high accuracy provides a useful tool for identifying students at risk of procrastination.

This study provides a robust predictive model for academic procrastination among college students, particularly highlighting the importance of subjective well-being, smartphone addiction, and negative emotions. The findings offer valuable insights for educators and institutions to design interventions focusing on promoting positive mental health, managing technology use, and addressing emotional challenges. While other factors also play a role, the model demonstrates the key predictors and their relative importance. Future research could examine longitudinal effects, explore the model's generalizability across different contexts and populations, and investigate additional factors to further refine the predictive accuracy.

This study's cross-sectional design limits the understanding of causal relationships and the dynamic interplay of factors over time. The sample was drawn from a single university in one region of China, which may limit the generalizability of the findings to other contexts. Future research should employ longitudinal studies and broader geographical sampling to address these limitations. Additionally, the study's reliance on self-reported data may be subject to social desirability bias.