05 / 2026 / Data Science / ML
Early Warning System
Online learning teams need earlier risk signals from VLE activity and assessment behavior so tutor intervention can happen before failure or withdrawal.
Client
Context and business domain
00 / Design System
Palette Logic
Data Science / ML
05
2026
Typography
Inter
Aa Bb Cc Dd Ee Ff Gg Hh Ii Jj Kk Ll Mm Nn Oo Pp Qq Rr Ss Tt Uu Vv Ww Xx Yy Zz 0123456789
Layout Grid
12-column responsive index structure.
Deep Slate
#0D1B1E
Graphite
#3D3D3D
Clear White
#F8F8F8
Soft Signal
#C9FBC6
Connected System
Predictive Analytics connects the visual language, interaction rhythm, and evidence structure for this case study.
01 / Challenge
The Challenge
Online learning teams need earlier risk signals from VLE activity and assessment behavior so tutor intervention can happen before failure or withdrawal.
02 / Methodology
Methodology / Experiment
- 01
Data Lifecycle
Processed OULAD raw data through cleaning, integration, feature engineering, modeling, and dashboard visualization.
Data pipeline90% - 02
Model Comparison
Compared decision tree and logistic regression, then selected the more interpretable and slightly stronger model.
2 models93% - 03
Risk Dashboard
Translated predictive outputs into KPI tracking and intervention-oriented dashboard views.
Streamlit app84%
03 / Solution
The Solution
Risk Detection Dashboard
A Streamlit dashboard surfaces withdrawal risk, fail risk, engagement, assessment completion, and detection signals.
Intervention Evidence Model
Decision tree outputs make the strongest drivers easier to explain to academic support teams.
04 / Impact
Impact & Metrics
93.04%
Accuracy
The README reports decision tree accuracy for fail or withdrawal risk classification.
96.80%
Precision
The primary model prioritizes high-confidence risk identification.
89.77%
Recall
The model is designed to catch at-risk students early enough for support action.