Machine Learning Model Reliability & Monitoring Dashboard (Streamlit)

A production-style model monitoring dashboard that compares baseline vs production runs to track:

• Performance drift
• Data drift
• Prediction drift
• Data quality
• Model explainability (SHAP/LIME)

Supports Classification, Regression, NLP (sentiment + summarization), and Computer Vision workflows.

---

Overview

This project focuses on model reliability in production, going beyond training accuracy to continuously evaluate how models behave once deployed.
The dashboard enables users to select a model, production run, and analysis type to visually inspect degradation, drift, and quality issues.

---

Key Features

Performance Drift Analysis

• Classification: F1, Precision, Recall, ROC-AUC, Accuracy, Confusion Matrix, PR Curve
• Regression: R², RMSE, MAE, MAPE, Predicted vs Ground Truth plots
• NLP: BLEU, ROUGE, similarity metrics and distributions

Data Drift Analysis

• Numeric features: Kolmogorov–Smirnov (KS) Test
• Categorical features: Chi-Square Test
• Feature-wise drift detection and visualization
• NLP-specific syntax and semantic drift checks

Prediction Drift Analysis

• Tracks changes in prediction distributions across production runs
• Statistical tests and trend plots (mean, median, max, min)

Data Quality Analysis

• Completeness (missing values)
• Uniqueness (duplicate detection)
• Validity (datatype mismatch, invalid values, outliers)
• Embedded dataset profiling using ydata-profiling

Model Explainability

• Tabular models: SHAP (waterfall, bar, violin, force, scatter plots)
• NLP models: LIME-based text explanations
• CV models: LIME explanations highlighting influential image regions

---

Tech Stack

• Python
• Streamlit
• Pandas, NumPy
• Scikit-learn
• SHAP, LIME
• Plotly, Matplotlib
• TensorFlow / Keras (Computer Vision)
• OpenCV
• NLP utility libraries

---

Project Structure

.
├── 1_Landing_Page.py
├── pages/
│   ├── 2_Analysis_Page.py
│   ├── Add_Model.py
│   ├── models/
│   │   └── <ModelName>/
│   │       ├── baseline.csv
│   │       ├── model.joblib / model.h5
│   │       ├── Production Runs/
│   │       ├── Ground Truths/
│   │       └── ...
├── utils.py
├── requirements.txt
└── README.md

---

Setup Instructions

1. Create and activate virtual environment

python -m venv .venv
# Windows
.venv\Scripts\activate
# macOS/Linux
source .venv/bin/activate

2. Install dependencies

pip install -r requirements.txt

3. Run the application

streamlit run 1_Landing_Page.py

---

Notes

• Some modules (e.g., NLP semantic drift embeddings, CV image paths) require dataset paths to match the repository structure.
• This project was rebuilt and modernized from an earlier academic implementation to reflect production-level monitoring concepts.

---

Future Improvements

• Fully configurable paths (no hard-coded directories)
• Dockerized deployment
• Alerting integrations (Slack / Email / Webhooks)
• Experiment tracking (MLflow or Weights & Biases)

---