THG CONSULTING GROUP
SSSP AI
Predictive
Modeling
PROJECT PROPOSAL
FEBRUARY 2024
1 REQUIREMENTS AND SOLUTIONS
Table
1.2 Background
1.3 Your Requirements
OF
1.4 Our Solutions
Contents 1.5 Our Team
2 PROPOSAL
2.1 Vision and Goals
2.2 Development
2.3 SSSP Student - Level Prototype
Deliverables and Scope
Success Criteria
Development Lines for future projects
1.2 Background
When the School & Student Success Platform
(SSSP) was in first development, the 1.4 Our solution
student level component was designed
to include a predictive model based
We propose to use bespoke AI and Machine
Summary, on historical data. After the Student
component was put on hold due to structural
changes at the company, this included all
Learning predictive modeling of the sort
used by insurance companies all over the
Requirements predictive components development. world to predict the probability of CSUSA
students requiring more or less support
& Solutions 1.3 Your requirements
CSUSA is in possession of years of valuable
from teachers. With these predictive
tools, teachers and instructors will have
client and student data that can provide
an insight into the future performance a greater insight into their students’
of its stakeholders. This data is an performance and will have in turn more
asset that is being underutilized and
maintained at great cost. CSUSA aims to information to create their Personalized
leverage this information to predict the Learning Plan (PLP) for each student,
potential performance of students, thereby
offering enhanced support to profiles that allowing them to adopt a modern preventive
may encounter greater difficulties in a perspective, as opposed to the traditional
proactive and ongoing manner throughout
the academic year. This will result in reactive approach. This will set CSUSA
improved student performance, leading to apart from competitors.
greater satisfaction among students and
their families, ultimately resulting in a
competitive business advantage.
Our Team
1.5
Rubén Herrera Giménez José Manuel González Eliel Tessone
Rodríguez
Rubén has years of experience as an Actuarial Jose has years of experience in Data Engineering, Eliel has been affiliated to CSUSA since 2017
Consultant at EY and Willis Towers Watson and he’s Machine Learning and Data Science in the fields working in the Department of Strategy under
a member of the Spanish Institute of Actuaries. of insurance, communication and construction. Erica Amorim and Zoie Saunders. He’s been an
Background He’s also a part time professor on the subject
of “Big Data in the Insurance Sector” within the
He has
infrastructure
built scalable
for data
and
pipelines
maintainable
in the
independent consultant since 2020, reporting to
Yolanda Thomas. His work experience with CSUSA
and experience Master’s program in Big Data and Data at UNED
(National Distance Education University).
cloud. Also, he has experience in microservices
and data-driven enterprise solutions.
and his knowledge of its business and company
needs will be crucial to reach CSUSA’s quality
standards.
He’s the lead author of two articles published
in international journals on Machine Learning.
Rubén has a degree in Business Administration from He has a PHD in Developmental Psychology and He has a bachelor’s degree in Economics and
the University of Valencia, with a specialization Machine Learning, a master’s degree in Educational Psychology from the George Washington University
in finance. Psychology and a second master’s degree in Data and a master’s degree in Data Science and Machine
Education and He has a master’s degree in Actuarial Science
from the University of Valencia and a master’s
Science and Machine Learning from the Complutense
University of Madrid.
Learning from Complutense University in Madrid.
qualifications degree in Data Science and Machine Learning from
the Complutense University in Madrid.
He’s also a PHD candidate in Quantitative Finance
and Actuarial Science from the University of
Valencia.
Rubén has participated in projects for the Jose has led different cross-functional Eliel has worked on the SSSP during its development
development of statistical and machine projects for data extraction using Computer and has regularly provided data updates during
learning models for calculating customer churn Vision, OCR and Text Recognition. He has also maintenance.
probabilities, risk valuation, and profit created infrastructure for cloud computing and
optimization. data pipelines, processing and transforming He has provided analysis and developed reports
Selected Projects He has
involving
extensive
the
experience
construction of
in projects
dashboards
data. He has experience using Clustering and
classification algorithms, and model prediction
for the stakeholder survey conducted every Fall
and Spring, as well as the new yearly Wellness
for stock storage. survey.
and interactive applications to present the
results of machine learning models and enhance
algorithm explainability for a better business
understanding.
Our Team Rubén Herrera Giménez José Manuel González
Rodíguez
Eliel Tessone
1.5
His experience in insurance His background as a Data His work experience with
as a whole, consulting and probability Engineer will allow raw data CSUSA and his knowledge of
key roles, modeling will ensure the
calibration and explanation
to be processed in the most its business and company needs
efficient manner, ensuring will be crucial to ensure that
and synergies of the AI solution, making its quality and enabling the our solution aligns with CSUSA’s
it more understandable for optimal performance of AI quality standards and work
education professionals. algorithms. philosophy.
- Probability modelling - Infrastructure & Data Pipelines
- Model explainability and - End to End Computer Vision solutions
interpretation of results
- ETL & data processing
- Interactive visualization
- Cloud services & Automation
Proposal
2.1 Further SSSP development of the student-level or less support from teachers (based on past
component should include an AI predictive algorithm performance and/or demographic information).
Vision and goals to provide the probabilities of CSUSA students
requiring more or less support from teachers.
Having this information at the beginning of the
year/semester will provide the teachers with an
This information will aid teachers in developing
additional tool when creating the personalized
a PLP for every student by giving them a greater
learning plans, as well as the classification
amount of actionable information.
of the most important features influencing the
Our goal, once the development is complete, is to probability of needing more or less assistance
provide teachers at the beginning of the year with from teachers and advisors.
the probabilities of CSUSA students requiring more
TEACHERS STUDENTS
INPUT WORK OUTPUT
Need for support
Development of PLP Improved academic performance
probabilities
EXAMPLE OF DEVELOPMENTAL STAGES (approximate, to provide a general idea of the entire process,
Proposal
2.2
pending discussion of the detailswith CSUSA):
1.-INPUT: 4.-INTEGRATION:
Development Historical databases of students with academic
and non-academic data.
The AI solution will be integrated into SSSP 2.0
so that at the beginning of the course, teachers
can use these probabilities to proactively tailor
2.-DEFINING OUR TARGET VARIABLE the PLP for each student.
Our team will work closely with Yolanda (NEED FOR SUPPORT):
Based on past grades, we will create three levels 5.-OUTPUT:
Thomas and Boris Shemigon to develop
(Intense support need from the letter grade D+ As an example: a student, based on their
and integrate this component into the characteristics, would obtain the following
to F, Additional support need from B-to C-, and
development of SSSP 2.0 beginning in probabilities:
Basic support need from A+ to B).
February 2024 in time for deployment in
September of 2024. These dates 3.-CALIBRATION OF AN AI SOLUTION: Intense support need: 60%
are approximate and will be confirmed Using the aforementioned target variable and
Additional support need: 20%
once certain technical aspects regarding student characteristics, we will develop an AI
In this case, the student has a high probability of needing
the input data, such as its structure, model capable of obtaining the probabilities intense support from the teacher, so the teacher should adapt
the PLP accordingly.
granularity, quality, and collection for each student to belong to each of the three
groups. Basic support need: 20%
process, have been analyzed.
As teachers log in to the SSSP and click on their classes, they will be able to see the predicted levels
of support needed for every student, color coded, along with all other necessary data to build every
personalized learning plan. The teachers would then complete the PLP column themselves.
SSSP Mr. Smith's 4th Grade Math Class
Student- Student ID Student Name GPA NWEA mid NWEA end
2022-2023
Math Grade
2023-2024
Math Grade
2024-2025
PLP
Level
Predicted Tier
Actual Actual
111111111 Beth Wise 2.85 291 200 D C Tier 2 Additional support - 1 hour per week
111111112 Neha Rivers 1.68 287 261 F D Tier 3 Intense support - 3 hours per week
Prototype
111111113 Sumaya Miles 2.16 277 222 C C Tier 3 Intense support - 2 hours per week
111111114 Rhonda Ramsey 2.09 219 238 C D Tier 3 Intense support - 3 hours per week
111111115 Evan Glover 4.00 295 269 A A Tier 1 No support
111111116 Deborah Fernandez 3.87 224 216 B A Tier 1 No support
2.3
111111117 Anisa Calderon 3.52 289 226 B B Tier 1 Basic support - 10 minutes per week
111111118 Annabel Adkins 3.77 298 214 A B Tier 2 Additional support - 1 hour per week
111111119 Darcey Russo 2.96 271 201 A C Tier 2 Additional support - 1 hour per week
111111120 Dominik Mccarthy 3.05 204 254 D B Tier 2 Additional support - 30 minutes per week
Label
Tier 1 Independent - Basic support needed
Tier 2 Developing - Additional support needed
Tier 3 Needs Improvement - Intense support needed
Proposal
3.3 Deliverables and Scope 3.4 Success Criteria 3.5 Development lines for future projects
Deliverables include the following: Success will be measured by the following. - Monitoring of results in the first full
year of algorithm implementation.
- Predicted algorithm integrated into - Acceptable metrics for the AI algorithm
SSSP platform. in classifying students into different - Review and periodic recalibration of
levels of support needs. For example, the algorithm to ensure it adapts to
- Predictions of academic success and the algorithm should be capable of new trends in success or failure.
failure probabilities, as an example. correctly classifying more than 75% of
case (indicative only to understand what
- Backtesting of the algorithm to the metrics refer to).
measure its prediction capability.
- Seamless integration into SSSP platform.
- Explanatory document of the algorithm
and development methodology.
THG CONSULTING GROUP
SSSP AI Predictive Modeling