🏛️ Uzbekistan Public Procurement Analytics

Vendor Risk Intelligence · Spend Optimization · Contract Clustering

---

📌 Project Overview

This project delivers a full data science pipeline applied to Uzbekistan's government procurement registry — specifically the direct contracts register of a public entity (STIR: 201122919) covering Q4 2025.

Government procurement is one of the most impactful yet underanalyzed data domains. This analysis transforms raw contract records into actionable intelligence for policymakers, auditors, and transparency researchers.

🎯 Business Questions Answered

#	Question	Method
1	Where does public money go?	Category-level Pareto analysis
2	Is there vendor concentration risk?	HHI index + Lorenz curve
3	Are there seasonal spending patterns?	Temporal trend analysis
4	Can contracts be segmented by risk profile?	K-Means + PCA clustering
5	What actions should procurement managers take?	Data-driven recommendations

---

📊 Dataset

Attribute	Value
Source	Uzbekistan E-Procurement Portal (xarid.uzex.uz)
Type	Direct Contracts Registry
Period	September 22 – December 24, 2025
Records	98 contracts
Columns	13 original + 8 engineered features
Entity	Single government buyer (STIR: 201122919)

Key Fields

contract_value  →  Contract amount in UZS
category        →  Procurement category (18 unique)
funding_source  →  Budget / Off-Budget / Reserve
vendor_id       →  Supplier tax ID (STIR)
delivery_days   →  Contractual delivery deadline
contract_date   →  Contract signing date

---

🔬 Methodology

Raw Excel Data
      │
      ▼
1. Data Cleaning & Normalization
   ├── Language normalization (Uz/Ru → Uz)
   ├── Column aliasing (PEP 8 names)
   └── Type casting & validation
      │
      ▼
2. Feature Engineering (8 new features)
   ├── log_value          → Handles value skewness
   ├── contract_month     → Seasonality signal
   ├── is_fast_delivery   → Urgency binary flag
   ├── spend_tier         → Quartile-based bucket
   └── funding_eng        → English funding label
      │
      ▼
3. Exploratory Data Analysis
   ├── Distribution & outlier analysis
   ├── Category Pareto (80/20 spend)
   ├── Temporal trend (monthly/quarterly)
   ├── Vendor concentration (HHI index)
   └── Correlation matrix + multivariate plots
      │
      ▼
4. Machine Learning — K-Means Clustering
   ├── Feature standardization (StandardScaler)
   ├── Optimal K via Elbow + Silhouette methods
   ├── PCA dimensionality reduction (2D projection)
   └── Cluster profiling & business labeling
      │
      ▼
5. Business Recommendations (5 actionable insights)

---

📈 Key Findings

💰 Spend Concentration — CRITICAL

• 1 vendor (200903001) accounts for ~89% of total spend in Waste Disposal
• HHI Score > 7,000 — classified as Highly Concentrated Market
• Top 3 vendors control >95% of all spend

🔍 Vendor Transparency Gap

• 11 contracts (11.2%) have anonymous vendor ID (X)
• All in Suvenir category — medium risk for compliance

📅 End-of-Year Spending Rush

• November–December show significant activity spike
• Classic sign of annual budget consumption pressure

🤖 ML Clustering Results (K=7, Silhouette-optimized)

Cluster	Profile	Dominant Category
0	High-value, known vendors	Waste Disposal
1	Low-value, fast delivery	Souvenirs
2	Medium-value, budget-funded	Hotel Services
3	Off-budget, rapid procurement	Catering
4	Micro contracts, anonymous vendors	Souvenirs
5	Utility contracts, long-term	Energy/Heating
6	Travel & transport cluster	Aviation

---

💡 Strategic Recommendations

Priority	Recommendation	Expected Impact
🔴 HIGH	Mandate competitive bidding > 5M UZS	Cost savings 10-20%
🟡 MED	Require STIR for all vendors	Compliance improvement
🟢 MED	Quarterly budget smoothing	Reduce year-end rush
🔵 MED	Cluster-based audit targeting	Audit efficiency +40%
🟣 LOW	Consolidate Suvenir purchases	Admin cost reduction

---

🖼️ Visualizations

Figure	Description
fig_01_distribution.png	Contract value distribution (raw + log + tier)
fig_02_category_spend.png	Pareto spend analysis by category
fig_03_funding_temporal.png	Funding structure + monthly trends
fig_04_vendor_concentration.png	Vendor HHI + Lorenz curve
fig_05_correlation.png	Correlation matrix + multivariate scatter
fig_06_optimal_k.png	Elbow method + Silhouette score
fig_07_pca_clusters.png	PCA 2D cluster visualization
fig_08_cluster_dashboard.png	Cluster profile comparison dashboard

---

🛠️ Tech Stack

Language   : Python 3.10+
Notebook   : Jupyter Notebook
Data       : pandas 2.x · numpy 1.x
ML         : scikit-learn (KMeans, PCA, StandardScaler, silhouette_score)
Viz        : matplotlib · seaborn
File I/O   : openpyxl

---

🚀 How to Run

Prerequisites

pip install pandas numpy matplotlib seaborn scikit-learn openpyxl jupyter nbformat

Clone & Run

# 1. Clone the repository
git clone https://github.com/thed700/uzbekistan-procurement-analytics.git
cd uzbekistan-procurement-analytics

# 2. Place the dataset
mkdir data
# Copy procurement_data.xlsx into the data/ folder

# 3. Launch notebook
jupyter notebook uzbekistan_procurement_analytics.ipynb

# OR run all cells non-interactively
jupyter nbconvert --to notebook --execute uzbekistan_procurement_analytics.ipynb

Project Structure

uzbekistan-procurement-analytics/
├── data/
│   └── procurement_data.xlsx        ← Input dataset
├── uzbekistan_procurement_analytics.ipynb  ← Main notebook
├── fig_01_distribution.png
├── fig_02_category_spend.png
├── fig_03_funding_temporal.png
├── fig_04_vendor_concentration.png
├── fig_05_correlation.png
├── fig_06_optimal_k.png
├── fig_07_pca_clusters.png
├── fig_08_cluster_dashboard.png
└── README.md

---

👤 Author

Akmal Toshpulatov
Junior Data Analyst | Termiz, Uzbekistan
🔗 GitHub: thed700

Certifications: Kaggle Python · Pandas · Intro to SQL · Advanced SQL

---

📜 License

This project is licensed under the MIT LICENSE.
Dataset is sourced from Uzbekistan's public procurement portal and is publicly available.

---

"Data is the new oil — but only if you refine it."
This project demonstrates that even small government datasets can yield strategic intelligence when analyzed rigorously.