Leveraging Cross-Industry Data and Quantum-Inspired Feature Engineering for SME Supply Chain Finance

Guest Speaker : Chen Zhiming, Ph.D., Chief Operating Officer of Huace Data Science, edited by Pu Qihang, South China University of Technology, produced by DataFunTalk.

Guide : With the rapid development of financial technology, applying cross‑industry data to empower financial services faces challenges such as data opacity, isolation, and quality, especially in supply‑chain finance for small‑and‑medium enterprises (SMEs). Huace shares its digital supply‑chain finance practice and explores feature‑engineering techniques for cross‑industry data.

Challenges of using cross‑industry data

SME engine restoring the authenticity of enterprise operations

Graph‑based technology for detecting SME fraud

Quantum‑computing‑driven big‑data feature engineering

01 – Challenges of Using Cross‑Industry Data

In supply‑chain finance, many participants (core enterprises, upstream/downstream firms, supply‑chain companies) create data silos, making SME financing difficult due to lack of transparency and quality. Financial service providers must collect and process multi‑source data differently to assess SME creditworthiness.

1. Supply‑Chain Data References

Supply‑chain finance involves static data (e.g., business registration, litigation, shareholder information) and dynamic data (e.g., financing demand, interest rates, loan terms, partner characteristics).

2. Fusion Architecture

Static and dynamic data are processed separately in feature engineering to preserve their distinct characteristics. After individual processing, the resulting single‑dimensional vectors are concatenated and fed into a neural network for credit rating.

The diagram shows the data‑fusion process where H1‑H5 represent feature‑engineering stages and H6‑H8 represent neural‑network prediction stages.

3. Variable Importance

Variable‑importance analysis reveals that dynamic variables account for about 40% of the selected features, highlighting the value of incorporating both static and dynamic data.

Detailed feature processing improves model generalization by capturing diverse data signals.

02 – SME Engine Restoring Business Authenticity

The SME engine adjusts raw enterprise data using six years of industry experience and knowledge bases, improving the accuracy of business analysis and credit scoring.

1. Industry SME Experience + Big‑Data Risk Control

During operating‑analysis, due to data inaccuracies, the SME engine calibrates financial figures to reflect true enterprise conditions.

The engine combines six years of accumulated industry data, government‑sourced datasets, and proprietary knowledge bases to adjust outlier inputs.

2. Data and Model Aspects

Online micro‑loans rely on standardized, data‑driven risk control, integrating personal, corporate, and upstream/downstream data into a comprehensive assessment.

Note: IPC refers to on‑site investigation loan mode managed by a client manager.

The integrated credit‑rating model combines PBOC sub‑models, Huace’s data‑fusion scores, and third‑party data, achieving KS scores above 0.5 for SME loans.

3. Deep Industry Understanding

Credit limits are determined by sales, capital needs, debt ratios, and credit scores, focusing on restored and industry‑average data to reflect true financing needs.

4. Asset‑Quality Analysis – Predicting Pig‑Farm Loan Delinquency

By incorporating market price data (e.g., pig price, grain‑price ratio) into risk models, Huace improves delinquency prediction for agricultural loans.

03 – Graph‑Based Technology for SME Fraud Detection

Fraud detection combines core‑enterprise screening, big‑data verification, transaction history analysis, and a graph‑based engine that highlights suspicious patterns.

Two examples illustrate one‑degree and multi‑degree network analyses that uncover fabricated applications and hidden relationships.

04 – Quantum‑Computing‑Driven Big‑Data Feature Engineering

Huace explores Quantum‑Inspired Evolutionary Algorithms (QIEA) to enhance feature selection for credit models, leveraging quantum concepts on classical hardware.

1. Quantum‑Inspired Evolutionary Algorithm (QIEA)

QIEA integrates quantum state vectors and probabilistic gene swapping, improving over traditional Genetic Algorithms (GA) by preserving diverse, relevant features.

QIEA’s broader feature coverage leads to higher selection frequencies for many features compared to GA.

Performance evaluation shows QIEA maintains robustness under added Gaussian noise, with stability above 70% versus GA’s drop to ~50%.

05 – Q&A

Q: What is the relationship between technology and scenarios, and how should technology support future scenarios?

A: Technology must align with industry specifics; Huace builds industry knowledge bases (e.g., agriculture, new energy) to create reusable modules that enhance financial services.

Thank you for reading. Please like, share, and follow for more content.