========================================================
 L M S  |  Predictive Analytics & Insight Dashboard 
========================================================

Machine learning pipeline and lightweight analytics dashboard for Learning Management System (LMS) data. Includes feature engineering, model training (XGBoost + Random Forest baseline), evaluation artifacts, and a credential‑protected Streamlit interface.

Key Features

• Reproducible pipeline: collection → preprocessing → feature engineering → training → evaluation.
• Models: XGBoost (GPU aware) and Random Forest baseline.
• Feature engineering: statistical aggregates, ratios, temporal usage, optional PCA + ANOVA F‑test selection.
• Evaluation artifacts: confusion matrices, prediction distribution, feature importance.
• Dashboard: Streamlit with encrypted credential store (Fernet) for simple access control.
• Modular scripts enable selective re‑runs using cached intermediate CSV outputs.

Project Layout

Dataset/                   Raw LMS CSV inputs
data_collection.py         Merge & normalize source tables
data_preprocessing.py      Cleaning, encoding, scaling, train/test split
feature_engineering.py     Derived features (counts, temporal, ratios, PCA/ANOVA)
model_training.py          Train & persist XGBoost / RandomForest models
model_evaluation.py        Metrics + plots (confusion matrices, distributions)
streamlit_app.py           Authenticated analytics dashboard
insight_generation.py      Higher-level narrative insights (optional use)
generate_credentials.py    Create encrypted credentials store
credentials_encrypted.pkl  Encrypted user/password data
key.key                    Symmetric encryption key (keep secret!)
plots/                     Generated visualizations
model.xgb                  Saved primary model artifact
requirements.txt           Dependency lock snapshot
README.md                  Project documentation

Suggested future refinement (not applied):

src/
	data/
	features/
	models/
	api/
	dashboard/
tests/

Showcase Scope

This repository snapshot demonstrates:

• Data merging and preprocessing pipeline design.
• Feature engineering strategy for educational performance prediction.
• Comparative modeling (gradient boosting vs ensemble baseline).
• Artifact-driven evaluation (plots committed for review). Operational run instructions have been intentionally omitted to keep focus on structure and deliverables.

Visualizations & Evaluation

Core model evaluation artifacts (committed for reference):

Confusion Matrices

Overall	XGBoost	Random Forest

Feature Importance & Prediction Distribution

Feature Importance	Prediction Distribution

Interpretation (abridged): off‑diagonal confusion matrix cells indicate systematic errors; feature importance highlights leverage points; distribution plot surfaces class imbalance considerations.

Model & Evaluation Summary

Primary metrics: accuracy, precision/recall, confusion matrices, gain/impurity feature importance. Potential extensions (not implemented here): cross‑validation, hyperparameter optimization, SHAP explanations, calibration curves.

GPU Note

XGBoost configuration supports automatic GPU usage when an appropriate build is installed; otherwise falls back to CPU.

Credentials

Dashboard authentication uses an encrypted (Fernet) credential store suitable for lightweight demo access control.

Containerization

Minimal container assets illustrate packaging (not an active deployment pipeline):

Dockerfile (excerpt): Python 3.11 slim base, installs requirements, launches Streamlit on port 8501.

Build & run (optional):

docker build -t lms .
docker run --rm -p 8501:8501 lms

Or with compose:

docker compose up --build

Result: dashboard accessible at http://localhost:8501 .