Abstract:
This research study aimed at assessing the impact of data-driven predictive models in
minimizing student dropout rates in public higher learning institutions by using local datasets
from the Institute of Accountancy Arusha (IAA) as a case study. A case study research design
and a quantitative research approach were employed in this study. The study population
included students at IAA. The sample size was 386 respondents selected using random
sampling technique. Closed-ended and open-ended questionnaire, and Document analysis
were employed as methods for data collection. Descriptive analysis was used to analyse
responses from respondents. The analysis involved analyzing enrollment and graduands
statistics obtained from IAA examination offices to understand the trend of student dropout from
2020 to 2022. Subsequently, responses from students regarding factors contributing to student
dropout were collected and quantitatively analyzed to identify the most influential factors for
developing predictive models. Three distinct data-driven predictive models—Neural Network,
Logistic Regression, and Support Vector Machine—were developed, trained, tested, and
validated using information from graduated and dropped-out students between 2020 and 2022.
The Neural Network model exhibited superior performance, achieving an 86.6% prediction
accuracy compared to other models. The Neural Network model, utilizing information from
continuing students, identified that 14.7% of students were at risk of dropping out in the next
semester. This finding underscores the model's effectiveness in identifying at-risk students
based on local context datasets. Recommendations for implementation were suggested to IAA
and the Regulatory bodies, such as NACTVET and TCU, can use the findings to develop
supportive policies for widespread model adoption. Academic advisers benefit from informed
decision-making on student support programs, fostering evidence-based practices.
