Predictive Analytics Model for Student Dropout Risk in Tertiary Institutions

Background to the Study

Student attrition — withdrawal of student from tertiary academic programme prior to completion whether through voluntary withdrawal academic exclusion or extended ultimately unresumed leave of absence — represents substantial and persistent challenge for higher education institutions globally. Beyond personal and economic cost borne by affected student whose interrupted education may leave them with accumulated debt or foregone earnings without corresponding qualification student attrition represents direct institutional cost: resources invested in student's recruitment enrolment and early academic support are not recouped through eventual graduation and elevated attrition rates can affect institution's funding accreditation standing and reputation.

Institutional responses historically predominantly reactive: academic advising and support interventions frequently triggered only after student already exhibited overt signs of academic distress (failing grades extended absence formal withdrawal application) by which point range of effective intervention options may be substantially narrowed relative to what might have been possible with earlier identification. Educational data mining and learning analytics research has over past decade increasingly explored whether routinely collected institutional data — demographic information captured at admission socioeconomic indicators such as tuition payment status and scholarship/loan status and early academic performance records — can be used to construct predictive models capable of identifying individual students at elevated dropout risk substantially earlier in academic programme than conventional reactive advising practice typically allows creating window for proactive rather than reactive institutional intervention.

Machine learning-based predictive analytics offers natural technical approach: given historical dataset of past students demographic socioeconomic and academic performance records together with eventual outcome (graduated still enrolled or dropped out) classification model can be trained to learn statistical patterns associated with eventual dropout and subsequently applied to current students records to generate individual dropout risk estimate. Practical value depends critically on ability to generate sufficiently accurate predictions using only information available early enough in student's programme for meaningful intervention to remain possible — constraint this study addresses directly by restricting model input features to information available by end of student's second semester.

This study designs implements and evaluates predictive analytics model for estimating individual student dropout risk in tertiary institutions comparatively evaluating multiple classification algorithms examining which student characteristics are most predictive of dropout risk and considering practical institutional deployment implications including fairness considerations relevant to its responsible use as decision-support tool for academic advising rather than punitive screening mechanism.

Statement of the Problem

·         Tertiary institutions frequently identify at-risk students only after overt academic distress already emerged narrowing window for effective proactive intervention.

·         While routinely collected institutional data (demographic socioeconomic and early academic performance records) contains statistical signal predictive of eventual dropout this data frequently not systematically analysed or operationalised into early-warning system supporting academic advising practice.

·         Published dropout prediction studies frequently do not explicitly restrict model input features to information available early enough in student's programme for resulting prediction to be practically actionable limiting real-world deployability.

·         Limited published research addressing dropout risk prediction within Nigerian tertiary institutions specifically and limited attention in broader predictive analytics literature to fairness implications of deploying such models as institutional decision-support tools.

·         Without empirically validated feature importance analysis institutions lack clear evidence on which specific student support interventions (academic financial or advising-related) most likely to address underlying risk factors identified by predictive model.

Aim and Objectives of the Study

Aim is to design implement and evaluate predictive analytics model for estimating individual student dropout risk in tertiary institutions using data available early enough in academic programme to support proactive institutional intervention.

·         Review existing student attrition theory and predictive analytics approaches to dropout risk estimation

·         Design and implement data preprocessing and feature engineering pipeline restricting model inputs to information available by end of student's second semester

·         Design train and comparatively evaluate four classification algorithms — Logistic Regression Random Forest XGBoost and Deep Neural Network — for dropout risk prediction

·         Conduct feature importance analysis to identify which demographic socioeconomic and academic performance factors are most predictive of dropout risk

·         Evaluate precision-recall trade-off at varying risk-classification thresholds relevant to institutional resource allocation for intervention

·         Conduct subgroup performance comparison to assess whether model's predictive performance consistent across relevant demographic subgroups as basic fairness check

Research Questions

·         What predictive accuracy precision recall F1-score and AUC are achieved by Logistic Regression Random Forest XGBoost and DNN models for dropout risk prediction using only second-semester-and-earlier data?

·         Which demographic socioeconomic and academic performance features are most predictive of dropout risk?

·         How does precision-recall trade-off vary across different risk-classification thresholds and what are implications for institutional intervention resource allocation?

·         Does model's predictive performance vary meaningfully across relevant demographic subgroups raising potential fairness considerations for its institutional deployment?

Significance of the Study

Significant to tertiary institution academic support and advising offices seeking empirically grounded early-warning tool to identify at-risk students proactively potentially improving student retention outcomes and effectiveness of limited advising resources. Significant to broader educational data mining and learning analytics research community as methodologically transparent study combining multiple classification algorithms explicit early-prediction feature-availability constraints and subgroup fairness analysis within single consistent study approach less commonly reported together in existing literature. Significant to institutional policymakers considering ethical and fairness implications of deploying predictive analytics in academic advising contexts. Significant to researcher as practical demonstration of applied predictive analytics addressing socially consequential real-world problem.

Scope of the Study

Limited to design implementation and evaluation of predictive model using publicly available UCI/Kaggle “Predict Students' Dropout and Academic Success” dataset supplemented with synthetically constructed Nigerian-context dataset constructed to reflect institution-reported aggregate statistics given unavailability of comparable individual-level public Nigerian datasets. Study restricts model input features to information available by end of student's second semester consistent with early-warning use case and does not address later-programme dropout prediction using more complete academic history which discussed as future work. Study does not include live institutional pilot deployment or intervention-effectiveness evaluation which would require institutional partnership and longitudinal follow-up outside resource scope.

Limitations of the Study

·         Primary dataset originates from Portuguese higher education institution; while feature categories broadly generalisable specific distributions and relative predictive weighting may differ in Nigerian institutional context.

·         Supplementary Nigerian-context dataset synthetically constructed to reflect institution-reported aggregate statistics rather than drawn from individual-level live student records for data privacy and access reasons; findings specific to supplementary dataset should be interpreted as illustrative rather than validated on genuine individual-level Nigerian student data.

·         Study evaluates predictive accuracy and fairness at model level; does not evaluate downstream effectiveness of any specific intervention triggered by risk prediction which would require longitudinal intervention-based study design outside scope.

·         Restricting model inputs to second-semester-and-earlier data while methodologically appropriate for early-warning use case necessarily excludes later-programme information that might otherwise improve predictive accuracy representing deliberate accuracy-versus-actionability trade-off rather than oversight.

Operational Definition of Terms

Student Dropout: Withdrawal from tertiary academic programme prior to completion whether voluntary or through academic exclusion without subsequent resumption.

Predictive Analytics: Use of historical data and statistical/machine learning modelling to estimate likelihood of future outcome for given case.

Educational Data Mining: Application of data mining and analytics techniques to data generated within educational settings.

Early-Warning System: System designed to identify individuals at elevated risk of negative outcome early enough for preventive intervention to remain feasible.

Feature Importance: Measure of relative contribution of individual input feature to model's predictive performance.

Curricular Units Approved: Number of course/module credit units student successfully passed in given academic period used as academic performance feature.

Fairness (in predictive modelling): Property whereby model's predictive performance and error rates do not vary systematically in manner that disadvantages particular demographic subgroup.

Conclusion

XGBoost achieved strongest predictive performance accuracy 87.9% F1-score 0.869 AUC 0.932 followed by Random Forest 86.4% accuracy and DNN 85.1% accuracy with Logistic Regression trailing 81.2% accuracy but providing directly interpretable coefficient-based risk factors. Feature importance analysis identified first- and second-semester grade performance number of enrolled versus approved curricular units age at enrolment and tuition payment status as most predictive features with socioeconomic and financial features collectively contributing substantial share of predictive power alongside purely academic indicators. Supplementary threshold-sensitivity analysis examined precision-recall trade-off at varying risk-classification thresholds relevant to institutional decisions about intervention resource allocation. Study concludes machine learning-based dropout risk prediction using data already routinely collected by tertiary institutions can identify at-risk students with sufficient accuracy and lead time to support proactive institutional intervention and recommends institutional pilot deployment as early-warning decision-support tool integrated with existing academic advising workflows with appropriate attention to fairness and non-punitive use of risk predictions.

Frequently Asked Questions (FAQs)

1. Why predict dropout by end of second semester?

Any useful early-warning must generate predictions early enough for meaningful intervention. Restricting inputs to second-semester-and-earlier data ensures window for proactive advising financial aid counselling or remediation rather than reactive action after withdrawal.

2. Which model performed best?

XGBoost 87.9% accuracy F1 0.869 AUC 0.932, followed by Random Forest 86.4% DNN 85.1% Logistic Regression 81.2% interpretable baseline providing directly interpretable coefficients.

3. What data were used?

UCI/Kaggle Predict Students' Dropout and Academic Success dataset 4,424 records Portuguese institution demographic socioeconomic first and second semester academic features plus synthetically constructed Nigerian-context dataset 2,000 records reflecting aggregate statistics from three Nigerian universities given unavailability of individual-level public Nigerian data.

4. What features are most predictive?

First and second semester grade performance number of enrolled versus approved curricular units age at enrolment and tuition payment status most predictive; socioeconomic and financial features collectively substantial share alongside purely academic indicators.

5. How does threshold choice affect intervention?

Threshold-sensitivity analysis examined precision-recall trade-off: lower threshold captures more at-risk students higher recall but more false positives requiring more advising resources; higher threshold higher precision fewer resources but may miss some at-risk. Institutional resource allocation decision.

6. Was fairness assessed?

Yes subgroup performance comparison assessed whether predictive performance consistent across relevant demographic subgroups as basic fairness check ensuring model not systematically disadvantaging particular subgroup when deployed as advising decision-support not punitive screening.

7. What are limitations?

Primary dataset Portuguese feature distributions may differ Nigerian context, Nigerian supplementary dataset synthetic illustrative not validated on genuine individual-level Nigerian records, study evaluates model accuracy not downstream intervention effectiveness requiring longitudinal design, second-semester restriction excludes later information deliberate accuracy-vs-actionability trade-off.

8. What is educational data mining?

Application of data mining analytics techniques to data generated within educational settings to uncover patterns predictive of outcomes such as dropout academic success engagement.

9. How does this differ from reactive advising?

Reactive advising triggered after overt distress failing grades extended absence withdrawal application narrowing intervention options; predictive analytics uses routinely collected data to identify elevated risk substantially earlier enabling proactive support.

10. Where download full project?

Download complete project with preprocessing feature engineering training and fairness analysis from SCHOLARNESTHUB as publication-ready document.