MBU Gap Analyzer

Predictive Gap Analysis Engine for Mandatory Biometric Updates

UIDAI Data Hackathon 2026 Submission

Problem • Solution • Features • Methodology • Visualizations • Installation • Tech Stack • Team

---

📋 Table of Contents

• Problem Statement
• Our Solution
• Key Features
• Deep Learning Edition
• Architecture
• Methodology Flow Diagram
• Key Findings
• Visualizations
• Installation
• Usage
• Dataset Structure
• Tech Stack
• Policy Recommendations
• Team

---

🎯 Problem Statement

The Hidden Crisis: Children Losing Benefits Due to MBU Non-Compliance

Children enrolled in Aadhaar are mandated to update their biometrics at two critical life stages:

Age	Update Type	Reason
5 years	Mandatory Biometric Update	Fingerprints mature, facial features change
15 years	Mandatory Biometric Update	Adolescent biometric changes

What Happens If They Don't Update?

❌ Aadhaar becomes INACTIVE
   ├── 🚫 School admission blocked
   ├── 🚫 Scholarship disbursement fails
   ├── 🚫 Mid-day meal authentication fails
   └── 🚫 DBT (Direct Benefit Transfer) denied

The Scale of the Problem

Thousands of children risk losing government benefits worth crores of rupees annually because their Aadhaar wasn't updated on time — often due to lack of awareness or inaccessible update centers.

---

💡 Our Solution

MBU Gap Analyzer: A 3-Layer Predictive Engine

We built an intelligent system that identifies "Ghost Cohorts" — children who were enrolled in Aadhaar but never completed their Mandatory Biometric Updates.

┌─────────────────────────────────────────────────────────────────┐
│                    MBU GAP ANALYZER                             │
├─────────────────────────────────────────────────────────────────┤
│                                                                 │
│   Layer 1: COHORT TRACKING                                      │
│   ├── Track children from enrolment → MBU age                   │
│   ├── Calculate MBU Compliance Ratio per district               │
│   └── Identify "Ghost Cohorts" with high gap                    │
│                                                                 │
│   Layer 2: SERVICE DESERT IDENTIFICATION                        │
│   ├── K-Means clustering on district performance                │
│   ├── Identify underserved areas                                │
│   └── Priority ranking for intervention                         │
│                                                                 │
│   Layer 3: SCHOLARSHIP RISK PREDICTION                          │
│   ├── Time series analysis of update trends                     │
│   ├── Forecast "Update Crunch" periods                          │
│   └── Predict children at risk before admission season          │
│                                                                 │
└─────────────────────────────────────────────────────────────────┘

---

✨ Key Features

📊 Cohort Tracking Aggregates 5M+ records across enrolment, biometric, and demographic data Computes district-wise MBU Compliance Ratio Risk classification: Green / Yellow / Red zones	🗺️ Service Desert Mapping K-Means clustering identifies underserved districts Silhouette analysis for optimal cluster selection Priority list for Aadhaar Seva Kendra deployment
📈 Predictive Forecasting 6-month ahead predictions using Prophet/Linear Regression Identifies "Update Crunch" before school admission season Confidence intervals for resource planning	💰 Impact Quantification Calculates scholarships at risk (Rs. Crore) DBT benefits potentially blocked Actionable financial impact for policymakers

---

🚀 Deep Learning Edition

Branch: deep-learning | Notebook: MBU_Gap_Analyzer_DeepLearning.ipynb

We've enhanced the standard ML solution with 4 cutting-edge Deep Learning modules to deliver a production-ready, hackathon-winning submission.

Module Overview

#	Module	Technology	Purpose
1	LSTM Forecaster	PyTorch	Time-series prediction for MBU demand during school admission rush
2	Autoencoder Anomaly	PyTorch	Unsupervised detection of suspicious districts with abnormal patterns
3	SHAP Explainability	SHAP Library	Explain WHY K-Means classified districts as Service Deserts
4	Geospatial Map	Folium	Interactive India map with color-coded Service Desert markers

🧠 Module 1: LSTM Time-Series Forecaster

Architecture: Input → LSTM (2 layers, 64 hidden) → FC → Output

LSTM (Long Short-Term Memory) captures temporal patterns in biometric update trends to predict future MBU demand.

$$h_t = o_t \odot \tanh(C_t)$$

• Training: 100 epochs with Adam optimizer + learning rate scheduling
• Output: 6-month forecast to anticipate school admission rush (June-July 2026)
• Use Case: UIDAI can pre-deploy mobile Seva Kendras to high-demand districts

🔍 Module 2: Autoencoder Anomaly Detector

Encoder: 5 features → 32 → 16 → 8 (bottleneck)
Decoder: 8 → 16 → 32 → 5 features (reconstruction)

Autoencoder learns normal patterns and flags districts with high reconstruction error as anomalies.

$$\text{Anomaly if } \mathcal{L}_{reconstruction} > \mu + 2\sigma$$

• Detection: Districts with abnormal enrolment-to-update ratios
• Use Case: Identify data quality issues or potential fraud

📊 Module 3: SHAP Explainable AI

SHAP (SHapley Additive exPlanations) uses game theory to explain model decisions.

• Method: KernelSHAP for K-Means clustering
• Output: Feature importance showing why a district is classified as Service Desert
• Use Case: Provide transparent, auditable explanations for policy decisions

📊 Deep Learning Visualizations

Model Benchmarking: LSTM vs Baselines

Key Result: LSTM reduces RMSE by 1.7% vs Linear Regression and 17% vs Moving Average

LSTM Training Progress

Training converges smoothly over 100 epochs with learning rate scheduling

LSTM 6-Month Forecast

LSTM predicts MBU demand surge during school admission season (June-July 2026)

🗺️ Module 4: Interactive Geospatial Map

Folium generates an interactive HTML map of India with:

Marker Color	Status	Compliance
🔴 Red	Service Desert	< 50%
🟠 Orange	At Risk	50-80%
🟢 Green	Compliant	> 80%

• Popups: District name, compliance ratio, MBU gap
• Output: analysis_outputs/service_desert_map.html

Quick Start (Deep Learning Edition)

# Switch to deep-learning branch
git checkout deep-learning

# Install additional dependencies
pip install torch shap folium

# Run the notebook
jupyter notebook MBU_Gap_Analyzer_DeepLearning.ipynb

---

🏗️ Architecture

Core Algorithm: MBU Compliance Ratio

$$\text{MBU Compliance Ratio} = \frac{\text{Biometric Updates (Age 5-17)}}{\text{Eligible Population (Enrolments Age 0-5 + Age 5-17)}} \times 100$$

Risk Classification Matrix

Compliance Ratio	Risk Category	Action Required
≥ 80%	🟢 Green (Compliant)	Maintain current operations
50% - 80%	🟡 Yellow (Moderate Risk)	Awareness campaigns needed
< 50%	🔴 Red (Service Desert)	Urgent intervention required

Machine Learning Pipeline

Input Data                    Processing                      Output
─────────────────────────────────────────────────────────────────────
                             ┌──────────────┐
[Enrolment CSV] ────────────►│              │
                             │   Merge &    │     ┌─────────────────┐
[Biometric CSV] ────────────►│  Aggregate   │────►│ Gap Analysis DF │
                             │   by Dist.   │     └────────┬────────┘
[Demographic CSV] ──────────►│              │              │
                             └──────────────┘              ▼
                                                  ┌─────────────────┐
                                                  │ StandardScaler  │
                                                  └────────┬────────┘
                                                           ▼
                                                  ┌─────────────────┐
                                                  │ K-Means (k=4)   │
                                                  └────────┬────────┘
                                                           ▼
                                              ┌────────────────────────┐
                                              │  Cluster Assignment    │
                                              │  • Service Desert      │
                                              │  • At Risk             │
                                              │  • Moderate            │
                                              │  • High Performer      │
                                              └────────────────────────┘

---

📊 Key Findings

1,070+ Districts Analyzed

177,900 Children at Risk

147 Service Desert Districts

Rs. 32+ Cr Benefits at Risk

District Risk Distribution

Green (Compliant)     ████████████████████████████████████████  930 districts (87%)
Yellow (Moderate)     ██                                         32 districts (3%)
Red (Service Desert)  ██████                                    147 districts (14%)

---

📈 Visualizations

Our analysis generates 8 interactive visualizations for the hackathon submission:

#	Chart	Purpose
1	Problem Districts by State	Identify states with most Service Desert + Moderate Risk districts
2	MBU Gap Treemap	Visualize children at risk by state (size = gap, color = severity)
3	MBU Gap by State	Children at risk of losing benefits per state
4	Optimal Cluster Selection	Elbow Method + Silhouette Score for K-Means tuning
5	Service Desert Scatter Plot	K-Means clustering with district-level risk categories
6	Daily Update Trend	Time series with 7-day moving average
7	Forecast Chart	Prophet-based 6-month prediction for school admission season
8	Financial Impact Bar	Scholarships & DBT benefits at risk (Rs. Crore)

Chart Gallery

1. Problem Districts by State

2. MBU Gap Treemap

3. MBU Gap by State (Children at Risk)

4. Optimal Cluster Selection (Elbow + Silhouette)

5. Service Desert Identification (K-Means Clustering)

6. Daily Biometric Update Trends

7. MBU Forecast: Predicting the Update Crunch

8. Financial Impact of MBU Non-Compliance

📁 All visualizations are also available as interactive HTML files in analysis_outputs/

🗺️ Interactive Service Desert Map

🔗 Click the map above or here to view the Interactive India Map

Explore the Folium-powered geospatial visualization showing Service Desert districts across India:

• 🔴 Red markers = Service Desert districts (< 50% compliance)
• 🟡 Yellow markers = Moderate Risk districts (50-80%)
• 🟢 Green markers = Compliant districts (> 80%)
• Click markers for district-level details (MBU Gap, Compliance %)

---

🚀 Installation

Prerequisites

• Python 3.10 or higher
• pip package manager
• Git

Quick Setup

# 1. Clone the repository
git clone https://github.com/STIWARTs/UIDAI_DH_2k26.git
cd UIDAI_DH_2k26

# 2. Create virtual environment
python -m venv .venv

# 3. Activate virtual environment
# Windows (PowerShell)
.venv\Scripts\Activate.ps1
# Windows (CMD)
.venv\Scripts\activate.bat
# Linux/Mac
source .venv/bin/activate

# 4. Install all dependencies
pip install -r requirements.txt

What's Included in requirements.txt

Category	Packages
Core Data Science	numpy, pandas, scipy
Machine Learning	scikit-learn, shap
Deep Learning	torch (PyTorch)
Time Series	prophet
Visualization	matplotlib, plotly, folium, kaleido
Jupyter	ipykernel, ipython, jupyter_client

⚡ Note: The full installation may take 5-10 minutes due to PyTorch and Prophet dependencies.

Deep Learning Edition Dependencies

pip install torch shap folium kaleido

---

📖 Usage

Run the Analysis

Standard ML Edition (MAIN branch)

1. Open Jupyter Notebook

   jupyter notebook MBU_Gap_Analyzer.ipynb

2. Execute All Cells

   • Press Shift + Enter to run cells sequentially
   • Or use Cell → Run All from menu

3. View Outputs

   • Interactive charts display inline
   • CSV exports saved to analysis_outputs/ folder
   • HTML charts for PDF conversion

🚀 Deep Learning Edition (deep-learning branch)

1. Switch to deep-learning branch

   git checkout deep-learning

2. Install Deep Learning dependencies

   pip install torch shap folium

3. Open the Deep Learning notebook

   jupyter notebook MBU_Gap_Analyzer_DeepLearning.ipynb

4. Execute All Cells - Includes 4 advanced AI modules:

   • Module 1: LSTM Time-Series Forecaster
   • Module 2: Autoencoder Anomaly Detector
   • Module 3: SHAP Explainable AI
   • Module 4: Folium Geospatial Map

Output Files

analysis_outputs/
├── mbu_gap_analysis_by_district.csv    # Full gap analysis
├── state_wise_compliance_summary.csv   # State-level metrics
├── service_desert_districts.csv        # Priority intervention list
├── ghost_cohort_districts.csv          # Top 20 underperforming districts
├── chart1_state_compliance.html        # Interactive visualizations
├── chart2_gap_treemap.html
├── chart3_service_desert_scatter.html
├── chart4_daily_trends.html
├── chart5_forecast.html
├── chart6_financial_impact.html
└── service_desert_map.html             # 🆕 Interactive Folium map (Deep Learning Edition)

---

📁 Dataset Structure

uidai_datasets/
├── api_data_aadhar_enrolment/          # ~1M records
│   ├── api_data_aadhar_enrolment_0_500000.csv
│   ├── api_data_aadhar_enrolment_500000_1000000.csv
│   └── api_data_aadhar_enrolment_1000000_1006029.csv
│
├── api_data_aadhar_biometric/          # ~1.8M records
│   ├── api_data_aadhar_biometric_0_500000.csv
│   ├── api_data_aadhar_biometric_500000_1000000.csv
│   ├── api_data_aadhar_biometric_1000000_1500000.csv
│   └── api_data_aadhar_biometric_1500000_1861108.csv
│
└── api_data_aadhar_demographic/        # ~2M records
    ├── api_data_aadhar_demographic_0_500000.csv
    ├── api_data_aadhar_demographic_500000_1000000.csv
    ├── api_data_aadhar_demographic_1000000_1500000.csv
    ├── api_data_aadhar_demographic_1500000_2000000.csv
    └── api_data_aadhar_demographic_2000000_2071700.csv

Data Schema

Dataset	Columns
Enrolment	date, state, district, pincode, age_0_5, age_5_17, age_18_greater
Biometric	date, state, district, pincode, bio_age_5_17, bio_age_17_
Demographic	date, state, district, pincode, demo_age_5_17, demo_age_17_

---

🛠️ Tech Stack

Python 3.10+ Core Language	Pandas Data Processing	NumPy Numerical Computing	Plotly Visualizations	Scikit-Learn ML Clustering	PyTorch Deep Learning
SHAP Explainable AI	Folium Geospatial Maps	Prophet Time Series	Kaleido Chart Export	Jupyter Notebook Interactive Development

---

📋 Policy Recommendations

Based on our analysis, we recommend the following interventions:

#	Recommendation	Impact
1	Deploy Mobile Aadhaar Seva Kendras in Service Desert districts	Reach underserved rural/tribal areas
2	Launch SMS/IVR Reminder Campaigns for children turning 5/15	Proactive awareness
3	Integrate MBU Status Check with school admission portals	Flag inactive Aadhaar early
4	Extend Fee Waiver beyond Oct 2025 for low-compliance states	Remove financial barrier
5	Deploy Real-time Dashboard for District Collectors	Weekly progress monitoring

---

👥 Team

Team OMEGA

UIDAI Data Hackathon 2026

Name	Role	LinkedIn
Piyush Verma	Team Leader
Stiwart Stance Saxena	Team Member

---

📄 License

This project is submitted as part of the UIDAI Data Hackathon 2026. All rights reserved.

---

Built with ❤️ for Digital India

_{Making Aadhaar work for every child}