Mining quantitative & qualitative data for insights on improvement of student performance The research population consists of students from RIT Dubai who started their undergraduate studies at RIT Dubai in 2014, 2015, 2016 and 2017.
The research population size used was different for academic data and for non-cognitive data.
Academic data
The academic data contained 213 rows, which represent 213 undergraduate student records. The 31columns represent 31 attributes of each record. The attributes are a mix of numeric and character datatypes. Academic data collected from RIT had high school scores, placement exam scores and undergraduate scores.
Non-Cognitive data
Non- Cognitive skills was measured in terms of 2 non-cognitive skills, Grit and Self Control.
The research population for the non-cognitive data were the respondents to the survey whose academic records were part of the academic data records used in the project.
The non-cognitive data set contained 67 records representing 67 such students
The academic data obtained from RIT needed cleaning, standardization and feature engineering before any analysis could be done.
Only relevant features were selected for analysis from the 31 features. Hence irrelevant features such as exam dates, High School names etc. were removed.
Data completeness: Missing values of numerical input variables were filled with the mean of each field in order to ensure data completeness.
The features used in analysis can be summarized as
Curriculum High school curriculum MathGrade High school Mathematics scores EnglishGrade High school English scores PhysicsGrade High school Physics scores MPE Mathematics placement scores PPE Physics placement scores EPE English language Proficiency scores GPA Cumulative GPA score from RIT GritScale Grit calculated from survey results SelfControlScale Self-Control calculated from survey results
Findings The study found that students from different curricula perform differently at RIT Dubai. This was validated with differences in their GPA variabilities. According to the population studied, Indian curriculum students, followed by African curriculum students, perform better than other curricula as the largest proportion of students scoring very high GPAs, students. Indian, African and British curriculum students demonstrate a consistent and stable performance over the first 3 years of study. Their placement scores also demonstrate better college readiness than other curricula as they have fewer students scoring lower than the average scores. The non-cognitive scales are also comparable although African curriculum students scored lower than Indian and British curriculum students.
American, SABIS and MOE curricula show drop in performance especially in the 3rd year of study. The reason for this needs to be studied with more records and student interviews, but one explanation could be the start of co-op causing time management issues. MOE and American curriculum students also are less college ready than others as found in the placement score analysis. This finding could be used by Academic student support at RIT and High Schools to support the students with Time management trainings and additional supported learning in Mathematics and Physics. There is a gap between the self-perception and reality in performance in American curriculum students as seen from the non-cognitive skill analysis. This needs to be taken into account while training these students.