The highest rate across all modalities. Private institutions: 29.5%. Public institutions: 21.3%. Table 1.
MASTER'S · MUST UNIVERSITY · FLORIDA, USA · 2025
The role of technologies in preventing school dropout.
A predictive analysis.
The master's thesis in which Marco investigates how machine-learning models can anticipate — and prevent — student dropout and financial delinquency in higher education before they happen.
Master of Science in Emergent Technologies in Education · Advisor: Prof. Dr. Michelle Patrícia Paulista da Rocha
THE QUESTION
What if technology could identify — and prevent — dropout before the student gives up?
The thesis starts from two structural problems that travel together in Brazilian higher education: school dropout and financial delinquency. One fuels the other. Payment difficulties push the student to quit; the quitting compromises the institution\'s sustainability. Marco investigates how predictive analytics can break that cycle before it closes.
Marco built the work on a bibliographic-documentary review. He did not design the experiment — he synthesized the state of the art and proposes a path for institutional application.
THE THESIS IN NUMBERS
Dropout in higher education is not uniform — and the data shows where it hurts most.
Four tables structure the bibliographic review. Each number below comes from a finding Marco consolidated from Teixeira, Mengtges & Kampf (2019), INEP (Censo da Educação Superior 2019) and Brdesee et al. (2022).
The field with the highest academic dropout, followed by Computer Science (30.5%). Health and Education stay below 22%. Table 2.
The factor that most drives dropout. Lack of adaptation to the course (21.3%) and low academic performance (16.7%) follow. Table 3.
Institutions that implemented predictive models reported a significant drop in dropout rates between 2015 and 2019. Table 4 — Brdesee et al. (2022).
METHODOLOGY
Bibliographic-documentary review with meta-analysis.
Marco mapped studies published between 2015 and 2024, with emphasis on the last five years — the period of greatest consolidation of emerging technologies in higher education. The collection combined specific descriptors across three major academic databases.
THE MODELS
The algorithms used to anticipate who is at risk.
Marco maps seven core machine-learning techniques applied to predicting school dropout and financial delinquency. Each with specific strengths — and limitations the thesis does not hide.
XGBoost
Extreme Gradient Boosting
Tree boosting. High accuracy on large datasets, handles categorical and numerical variables well, minimizes overfitting (Chen & Guestrin, 2016).
Random Forest
Multiple decision trees
Robustness on large educational datasets. Evaluates performance, attendance and LMS engagement, reducing overfitting (Sghir, Adadi & Lahmer, 2023).
SVM
Support Vector Machines
Particularly effective for dropout in distance-learning courses, where the lack of in-person interaction weighs more heavily (Tete et al., 2022).
GBM
Gradient Boosting Machine
Fine-tuning of hyperparameters to minimize statistical bias. Strong on seasonal dropout patterns.
KNN
K-Nearest Neighbors
Classifies students into risk categories with high accuracy based on comparable behavior (Trajano, 2023).
LSTM
Long Short-Term Memory
Deep learning for temporal data. Reached +4.61% in R² score by the fourth semester compared to traditional ridge regression (Brdesee et al., 2022).
SMOTE
Synthetic Minority Oversampling Technique
Not a predictive model: a technique to fix class imbalance. Raised AUC from 0.91 to 0.96 in a decision-tree model.
PREDICTIVE VARIABLES
The signals that precede the dropout.
The literature reviewed by Marco points to a set of behavioral, academic and financial variables that, combined, anticipate dropout and delinquency risk with increasing accuracy.
Academic
- Class attendance
- Performance in formal assessments
- Courses passed vs. failed
- Engagement in extracurricular activities
Behavioral
- LMS access and time of use
- Forum participation and online interactions
- Temporal engagement patterns (longitudinal series)
- Observable disengagement signals
Financial
- Late tuition payments
- Student financial history
- Estimated ability to pay
- Reliance on student credit or scholarship
Demographic and socioeconomic
- Family income and occupation
- Concurrent work hours
- Distance and mobility to campus
- Prior academic background
FROM PREDICTION TO ACTION
Prediction without action does not reduce dropout.
Chapter 4 of the thesis addresses the link that is usually missing: how to integrate predictive models with proactive interventions that truly change a student\'s trajectory. Marco points to four complementary fronts.
01
Proactive interventions integrated with the model
Mentoring, psychosocial support and academic incentives triggered by predictive alerts — not by the financial-loss history. Tete, Sousa, Santana & Fellipe (2022) show that this combination reduces dropout.
02
Personalized academic and financial support
Learning analytics + business intelligence applied to the individual journey. Recommendations for courses, materials and scholarships based on performance and estimated ability to pay.
03
AI in scholarship and funding management
Logistic regression and decision trees to predict payment capacity, prioritizing scholarships and flexible plans for high-risk-high-potential students (Lemos et al., 2017; Pimentel et al., 2024).
04
Fintech partnerships and flexible plans
Payment solutions designed with fintechs to reduce the financial barrier. Goldrick-Rab et al. (2016) show the direct impact of this type of assistance on completion rates.
FROM THESIS TO PRACTICE
Marco applies this thesis today as co-founder of Principia, the AI Operating System for Education: 760,000 students, 1,200 educational institutions and more than BRL 2 billion in credit portfolio under management. Predictive analytics leaves the page and becomes a daily decision lever.
See the work applied to credit-and-collections mentoring (Credi+ Club) →ETHICS AND GOVERNANCE
The thesis does not hide the hard part.
Predictive models in education carry risks that cannot be ignored. Marco devotes specific sections of the review to the frontier between analytical power and institutional responsibility.
Data privacy
Student academic and financial data demand rigorous protection. Transparency about use and storage is a precondition, not an accessory.
Algorithmic bias
Models can perpetuate educational inequalities if trained on biased datasets. Regular auditing and explainability are mandatory (Sghir, Adadi & Lahmer, 2023).
Transparency and explainability
Automated decisions must be justifiable to managers and educators. Bird (2023) underscores transparency as a condition for adoption.
Responsible governance
Predictive analytics does not replace human judgment — it informs it. Marco treats this as the central ethical boundary of the thesis.
Reading this chapter speaks directly to Marco\'s training as a Certified Board Member by Board Academy (PFCC, 2025): the frontier between AI and the fiduciary duty of the board.
FINAL CONSIDERATIONS
What the thesis sustains.
"The use of emerging technologies represents a significant advancement in educational management, promoting a more inclusive, sustainable, and data-driven academic environment. The adoption of such tools can transform institutional practices, contributing to improved student retention and the financial health of higher education institutions."
The research also highlighted the ethical and technical challenges associated with the use of student data, suggesting that the continuous improvement of these approaches can generate significant impact on student retention and on equity of access to higher education.
The thesis argues, in synthesis: early prediction + proactive intervention + responsible governance form the triad that can break the dropout cycle — and return financial predictability to institutions without losing the student along the way.
ACADEMIC CITATION
How to cite the thesis.
For use in articles, dissertations and professional references.
ABNT
SOUSA, Marco Antonio Cardoso de. The role of technologies in preventing school dropout: a predictive analysis. 2025. Final Conclusion Work (Master of Science in Emergent Technologies in Education) — MUST University, Florida, USA.
APA
Sousa, M. A. C. de. (2025). The role of technologies in preventing school dropout: A predictive analysis [Master's thesis, MUST University].
Keywords
Predictive Analytics, School Dropout, Financial Delinquency