A Dimensional student enrollment prediction model: case of Strathmore University
Alaka, Benard Ochieng
MetadataShow full item record
The rate of student admissions within most Kenyan Universities has thus far been met with a corresponding uncertainty in budgetary allocation. Additionally, the increase of most applicants not being enrolled has led to lower institution yield. Due to the uncertainty of the quantity of students to be enrolled, planning and budgetary issues have arisen as stated earlier. Departments in charge of recruiting students are left to speculate the numbers likely to turn up. This in most cases is not accurate since it results into gaps in the allocated budgets and straining of resources. Currently, in Kenya, there is no institutions of higher learning that has a reliable means of predicting the expected institutional yield. Rather, academic management systems exist and are used to manage daily academic routines. These systems are served by transactional databases which are subject to being edited frequently and as such lack the capability of archiving histories of instances of the data within these databases; which makes them unsuitable for carrying out analysis on enrollment prediction. As such, a dimensional enrollment prediction model is proposed so as to aid in forecasting; not only of how many admitted students will be enrolled but also particular individuals who could show up for the purposes of follow-up activities. The inputs to the proposed enrollment prediction system were sourced from dimensional data stored in a data warehouse regarding to student details as per the admission as well as snapshot data of third party satisfaction index from accredited sources. The proposed system then transforms this data into dimensional data by adding a time variant to it and then passing the data through a neural network. The resultant model is then to be used in predicting students’ enrollment. The proposed model was tested for accuracy using the precision, recall ratio and the F-score Measure. The model’s accuracy was considerably high with an accuracy of 71.39% with a precision of 0.72. The average recall ratio was 0.71 and while F-score of 0.71 as well was obtained. In relation to some of the works reviewed the proposed model was a bit lower accuracy due the dataset used that was noisy as fetched from real student transactional databases.