►
Description
April 2022
Improving College Success Through Predictive Modeling
Presenter: Mallory Sheff
Institution: DataKind
A
So
my
name
is
mallory
chef
and
I
have
been
working
with
datakind
for
the
past
two
years:
I'm
their
portfolio
manager
for
our
economic
resilience
body
of
work.
Under
this
portfolio
of
projects.
We
have
been
looking
at
four
key
thematic
areas
that
really
support
individual
and
community
resilience,
and
this
includes
post-secondary
education
as
a
real
cornerstone
of
general
resilience
in
the
united
states.
A
A
We
like
to
see
ourselves
as
a
bridge
where
we
bridge
the
gap
on
the
one
hand,
between
technologists
and,
on
the
other
hand,
on
the
social
sector,
so
we're
really
a
non-profit
that
connects
pro
bono
data
scientists
with
mission-driven
organizations
and
our
work
really
comprises
of
three
core
phases
of
every
partnership.
We
engage
in
to
make
sure
that
it
is
fully
collaborative
one
is
that
we
really
help
our
social
organizations
identify
their
data
and
ai
opportunities.
A
Where
are
their
pain
points
and
what
is
data
scienceable?
We
then
recruit
and
manage
a
team
of
pro
bono
experts
that
have
skills
that
perfectly
match
the
solution
that
we're
trying
to
build
together,
and
then
we
stick
around
to
ensure
that
the
syst,
the
solutions
are
really
sustainable
and
that
impact
is
in
fact
achieved
for
our
partners.
A
So
more
concretely,
what
does
this
mean
and
how
do
we
do
our
work?
So
I'm
now
really
excited
to
share
the
work
that
we've
done
in
partnership
with
john
jay
college
at
the
city
university
of
new
york
and
how
we
applied
our
data
science
strategies
to
help
university
students
graduate
with
their
bachelor's
degree.
A
A
A
Two-Thirds
of
college
dropouts
are
low-income
students
and
black
and
latino
students
are
disproportionately
affected,
and
while
we
have
to
acknowledge
that
college
completion
rates
have
been
inching
up
to
62
percent,
which
is
their
highest
level.
Yet.
As
per
a
2021
statistic,
we
know
that
this
is
just
still
not
enough.
A
A
A
Really
john
jay
wanted
to
understand
how
leveraging
existing
student
data
they
could
answer
a
set
of
questions.
For
example,
could
they
identify
students
who
were
likely
to
drop
out
to
preemptively
provide
support
and
what
were
some
of
the
factors
that
would
influence
the
student's
decision
to
leave
school
before
earning
their
degree?
A
So
using
the
available
data
that
was
provided
to
us
by
the
john
jay
college
team,
we
were
able
to
develop
these
models
to
predict
non-graduation
for
freshmen
transfer
students
within
the
cuny
system
and
transfer
students
outside
of
the
cuny
system.
So,
as
you
can
see
here,
we
have
three
separate
unique
models.
A
One
prediction
for
just
freshman
students,
one
prediction
for
transfers
within
the
cuny
system
and
one
prediction
model
for
students
who
are
outside
of
this
cuny
system,
and
we
were
then
able
to
aggregate
these
risk
scores
to
get
a
general
prediction
for
any
given
term
from
there.
We
were
able
to
develop
secondary
models,
so
the
information
that
was
generated
in
the
primary
set
of
models
really
informed
the
secondary
set
of
models.
A
These
models
are
actually
applied
for
different
terms,
but
every
single
student
freshman
transfers
within
the
cuny
system
and
transfers
from
outside
the
cuteness
system
are
combined
with
the
secondary
model.
The
john
jay
college
can
use
current
student
information
to
generate
and
update
a
student's
probability
of
not
graduating,
because
it
really
allows
for
changes
in
student
performance,
enrollment
status
and
other
factors.
A
A
So
as
an
example,
we
were
able
to
determine
that
a
student
who
has
an
average
gpa
is
completing
an
average
of
only
10
credits
per
term
and
has
failed
two
courses
in
the
past.
This
student
will
have
a
higher
probability
of
not
graduating,
and
each
of
the
models
that
I
presented
in
term
were
able
to
correctly
predict
the
students
that
will
graduate
with
80
to
90
accuracy
and
correctly
predict
students
who
will
not
graduate
with
about
70
to
80
accuracy,
which
is
a
huge
success.
A
However,
as
soon
as
we
were
able
to
hand
over
the
cusp
tool
for
the
john
jay
college
team
to
integrate
that
into
their
work,
they
were
able
to
support
87
percent
of
students
to
graduate
which
was
an
additional
600.
Students
were
able
to
complete
their
bachelor's
degree
and
the
cusp
tool
really
enabled
dara
and
her
team.
Who
is
our
project
champion
to
provide
students
with
key
support,
including
strategies
to
overcome
financial
and
academic
barriers
to
graduation?
A
A
So
we're
working
with
a
team
of
eight
pro
bono
data
scientists
to
drive
forward
on
the
following
milestones:
we're
working
on
exploratory
data
analysis
right
now
to
really
understand
the
data
that
we're
working
with
what
is
being
represented
in
the
data.
And
how
can
we
learn
from
the
data
we'll
be
driving
towards
a
minimum
viable
product?
So
really
developing
an
algorithm
or
a
predictive
model
that
the
john
jay
college
team
can
provide
us.
A
So,
as
a
quick
summary
of
our
collaboration,
we
were
able
to
build
predictive
models
and
really
enable
us
to
provide
data-driven
insights.
The
john
jay
college
team
on
students
that
are
at
highest
risk
of
dropping
out
and
not
completing
their
full
bachelor's
degree
we're
leveraging
insight
from
this
first
partnership
to
focus
specifically
on
transfer
students
who
were
identified
by
the
john
jay
college
team
as
being
an
at
risk
student
group
and
we're
really
leveraging
data-driven
information
to
see
how
we
can
empower
the
john
jay
college
team
to
preemptively
intervene
and
support
those
students.