Building Smart Student Engagement Detector: An AI-Powered...

Team members

This project was developed by:

Devendhar Rao @devendhar_rao
Madhan Chowdary @madhan_chowdary
Prabu Kiran @j_prabhukiran_9b653c71e
B. Rooprekha @roop_rekhabharde_eae873c
Srujana Sadhu @srujanasadhusharma
Chanda Raj Kumar Sir @chanda_rajkumar

We’d also like to thank @chanda_rajkumar Sir for the constant guidance and support throughout this project. A lot of the clarity we had in system design and implementation came from those discussions.

Why do students struggle without anyone noticing?

This is something we kept seeing again and again.
A student starts missing a few classes. They participate a little less. Their feedback becomes vague - “okay”, “fine”, nothing detailed.
Individually, none of this looks serious. But over time, it adds up.
And by the time it reflects in marks, it’s already late.
That’s where this idea came from - what if we could catch these signals early instead of reacting later?

What do we build?

We built a student engagement detection system that tries to answer one simple question:
“Is this student starting to disengage?”
Instead of depending on just marks or attendance, we combine multiple signals:

Academic performance
Attendance
Behavioral patterns
Written feedback The goal is not just prediction, but early awareness.

What does the system actually do ?

Once student data is available, the system:

Looks at attendance and marks
Processes feedback text using NLP
Combines everything into an engagement score
Classifies students as Engaged, Moderate, or At Risk
Shows the result along with a confidence score It’s not meant to replace teachers - just give them a clear early signal.

Tech stack

We kept things simple and practical:

Frontend: React + Tailwind CSS
Backend: Django Framework
Database: MongoDB
AI logic: ML + NLP
Visualization: basic charts for dashboards

Why MongoDB?

We went with MongoDB mainly because our data isn’t uniform.

A single student record can include:

Numbers (marks, attendance)
Text (feedback)
Computed results (scores, labels)

Trying to force all of that into rigid tables didn’t feel right.

MongoDB made it easier to:

Store mixed data
Update fields when predictions are generated
Fetch everything in one go for dashboards

AI / ML / NLP - what’s actually happening behind the scenes

We didn’t use anything overly complicated, but we focused on combining things properly.

Basic prediction model

At the core, we use a simple feature-based approach:
F = {attendance, marks, behavior, feedback}
Each of these contributes to the final engagement score.
For example:

Low attendance → increases risk
Low marks → increases risk
Negative feedback → strong signal It’s simple, but when combined, it becomes quite effective.

NLP for feedback analysis

This was one of the most useful parts.
Students often write things like:

“I didn’t understand this topic”
“This is confusing”
“It’s okay” Even if marks are fine, this kind of feedback can indicate a problem. So we use basic NLP to:
Detect sentiment
Identify confusion or negativity This adds a layer that numbers alone can’t capture.

Multimodal approach

Most systems look at one thing, marks or attendance.
We combine:

Numerical data
Behavioral data
Text data This gives a much more complete picture of what’s going on.

Deep learning (from research side)

In our research work, we also explored:

LSTM models for tracking patterns over time
Attention mechanisms to weigh features
Transformer-based NLP for deeper text understanding These aren’t fully implemented in the current system, but they show where this can go next.