SmartGPA: How Smartphones Can Assess and Predict Academic - PowerPoint PPT Presentation

SmartGPA: How Smartphones Can Assess and Predict Academic Performance of College Students Rui Wang, Peilin Hao, Xia Zhou, Andrew Campbell (Dartmouth College) Gabriella Harari (University of Texas at Austin)

performance

behaviors features correlations prediction level change

studentlife.cs.dartmouth.edu studentlife.cs.dartmouth.edu

we extend studentlife

behaviors features correlations prediction class attendance, studying and partying

semantics of location

studying

study studying areas focus activity sound

attending classes and studying 1 24 attendance study study duration (hours) 0.75 18 attendance 0.5 12 0.25 6 midterm 0 0 1 2 3 4 5 6 7 8 9 week

partying

party places partying sound activity co-location

partying trends across the term 6 party duration (hours) 4.5 3 1.5 mid term green key 0 1 2 3 4 5 6 7 8 9 10 week

party duration study duration 0.8 3.4 0.4 1.7 0 0 1 2 3 4 5 6 7 1 2 3 4 5 6 7 weekday weekday

behaviors features correlations prediction capturing the dynamics of behavior

how to represent the data? 5 study duration (hour) 3.75 2.5 1.25 0 1 2 3 4 5 6 7 8 9 week

use mean to measure level 5 study duration (hour) 3.75 mean = 3.15 2.5 1.25 0 1 2 3 4 5 6 7 8 9 week

behavior term slope 5 study duration (hour) 3.75 2.5 term slope = 0.29 1.25 0 1 2 3 4 5 6 7 8 9 week

behavior term slope 5 study duration (hour) 3.75 2.5 term slope = 0.29 1.25 midterm 0 1 2 3 4 5 6 7 8 9 week

pre/post midterm slope 14 pre-slope = 2.23 post-slope = -0.86 study duration 10 6 midterm 2 1 2 3 4 5 6 7 8 9 week

breakpoint — when students change their behavior to adapt 0.7 study duration (scaled) different breakpoint student 1 0.35 student 2 0 1 2 3 4 5 6 7 8 9 week

breakpoint — how to compute 4 study duration • iteratively select every week as breakpoint 2 • use one or two linear regressions to fit the data before and after the breakpoint 0 1 2 3 4 5 6 7 8 9 week

breakpoint — how to compute 4 study duration • iteratively select every week as breakpoint 2 • use one or two linear regressions to fit the data before and after the breakpoint 0 1 2 3 4 5 6 7 8 9 week

breakpoint — how to compute 4 study duration • iteratively select every week as breakpoint 2 • use one or two linear regressions to fit the data before and after the breakpoint 0 1 2 3 4 5 6 7 8 9 week

breakpoint — how to compute MSE 6 MSE 1 MSE 5 4 study duration 2 0 1 2 3 4 5 6 7 8 9 1 2 3 4 5 6 7 8 9 1 2 3 4 5 6 7 8 9 week week week we use Bayes Information Criterion to select the breakpoint

behaviors features correlations prediction which of the 193 features relate to performance?

studying, partying and GPA study duration study focus - activity study focus - audio party duration -0.45 -0.3 -0.15 0 0.15 0.3 0.45 0.6 R value

studying, partying changes and GPA pre-midterm class attendance pre-midterm study duration after-midterm conversation duration 0 0.113 0.225 0.338 0.45 R value

behaviors features correlations prediction what models can predict GPA?

studying w 0 partying w 1 use lasso to regularize training w 2 activity w 3 + conversation GPA … w i+1 stress / positive leave-one-out cross validation affect w i+2 mental health w i+3 personality

selected features three sensor-based behavioral features • conversation duration night breakpoint • conversation duration evening term-slope • study duration three EMA features • positive affect • positive affect post-slope • stress term-slope one personality • conscientiousness

prediction performance 1 goodness of fit: 0.75 • R 2 = 0.559 • r = 0.81, p < 0.01 CDF 0.5 our model can distinguish high and 0.25 lower performers MAE = 0.179 0 0 0.1 0.2 0.3 0.4 0.5 absolute error

Thanks, I’m done