01
Record
Users enter a small set of health and lifestyle fields, including age, height, weight, family history, water intake, and activity level.
Obesity Risk Prediction Showcase
A focused prototype for obesity risk prediction.
ELORA is a local-first web prototype that records a small set of lifestyle indicators and returns a clear obesity risk result in a simple, readable format.
Risk Assessment
Risk Assessment — the main result page used to present the final obesity risk outcome.
Core feature
One main function, presented directly.
This page presents the central workflow of the project: entering daily indicators, reviewing short-term change, and returning an obesity risk result.
02
Review
Saved entries can be viewed as recent weight and BMI trends so that short-term change is visible rather than hidden in isolated records.
03
Assess
The result page combines model output and a simple rule-based reference score into one hybrid risk view that is easier to interpret.
Working screens
Supporting screens from the implemented prototype.
These two supporting screens show the data entry view and the recent trend view that sit alongside the main assessment screen above.
Daily Record Input — the form used to enter the core health and lifestyle information before generating an assessment.
Recent Weight and BMI Trends — a simple visual summary of the latest saved records.
Result logic
A result view designed to be readable.
The assessment view separates overall evaluation, single-record certainty, and the final hybrid risk index so that the result is easier to read and explain.
Hybrid output
The final result is presented as a simple hybrid score. It combines prediction certainty for the current record with a rules-based reference score, making the result clearer for demonstration and discussion.
Hybrid Risk = 0.7 × prediction certainty + 0.3 × rules score
Offline evaluation snapshot
The prototype also keeps a separate offline evaluation summary. This helps distinguish one-record certainty from overall model performance on the dataset.
89.3%
Hold-out accuracy reported for the selected model.
88.1%
Macro-F1 used to reflect multi-class performance more fairly.