Engineered risk features for trade-exposure classification: a gradient boosting framework validated on synthetic firm data

Realignments in global trade relationships affect exporting firms unevenly, yet risk assessment commonly relies on sector aggregates that conceal firm-level differences in exposure and adaptive capability. Operational measures that combine trade exposure with supply-chain capability for individual firms remain scarce. This study examines whether a domain-engineered representation of firm-level trade vulnerability improves risk classification and whether the improvement depends on learner capacity. Five composite indicators (Trade Concentration, Export Vulnerability, Diversion Exposure, Resilience Gap and Import Dependency) are constructed from six trade and capability variables grounded in the Indian trade literature. The study employs a synthetic dataset of 10,000 firms with a specified risk structure, ten random seeds and an ablation of raw, engineered and combined feature sets across Decision Tree, Logistic Regression, Random Forest and Gradient Boosting classifiers. The results show that Gradient Boosting classifies firms into Low, Medium and High risk with 84.44 per cent accuracy, a macro-F1 of 0.8392 and one-versus-rest AUC values between 0.9168 and 0.9828; 99.74 per cent of misclassifications fall between adjacent categories. The macro-F1 gain from the engineered indicators declines from 0.0528 for the Decision Tree to 0.0370 for Logistic Regression, 0.0186 for Random Forest and 0.0062 for Gradient Boosting (Holm-adjusted p = 0.106). The five indicators alone perform within 0.0030 macro-F1 of the full eleven-variable set. The findings of this study suggest that domain feature engineering and model capacity act as partial substitutes and that export vulnerability, diversion exposure and import dependency are the dominant dimensions of firm-level trade risk within the benchmark. JEL Classification Codes: C45, C53, F14, F15, M11.

Authors

Institutions

Publication Details

Journal
American International Journal of Sciences and Engineering Research
Published
2026-11-13
DOI
https://doi.org/10.46545/fze96p33
Primary Topic
Global trade and economics
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

Engineered risk features for trade-exposure classification: a gradient boosting framework validated on synthetic firm data

Netranand Pathak, Shashikant Rai, Rhythm Rakesh, Chelpuri Sumithra Devi
1 links
American International Journal of Sciences and Engineering Research
Global trade and economics
article

Engineered risk features for trade-exposure classification: a gradient boosting framework validated on synthetic firm data

Netranand Pathak, Shashikant Rai, Rhythm Rakesh, Chelpuri Sumithra Devi
article en

Abstract

Realignments in global trade relationships affect exporting firms unevenly, yet risk assessment commonly relies on sector aggregates that conceal firm-level differences in exposure and adaptive capability. Operational measures that combine trade exposure with supply-chain capability for individual firms remain scarce. This study examines whether a domain-engineered representation of firm-level trade vulnerability improves risk classification and whether the improvement depends on learner capacity. Five composite indicators (Trade Concentration, Export Vulnerability, Diversion Exposure, Resilience Gap and Import Dependency) are constructed from six trade and capability variables grounded in the Indian trade literature. The study employs a synthetic dataset of 10,000 firms with a specified risk structure, ten random seeds and an ablation of raw, engineered and combined feature sets across Decision Tree, Logistic Regression, Random Forest and Gradient Boosting classifiers. The results show that Gradient Boosting classifies firms into Low, Medium and High risk with 84.44 per cent accuracy, a macro-F1 of 0.8392 and one-versus-rest AUC values between 0.9168 and 0.9828; 99.74 per cent of misclassifications fall between adjacent categories. The macro-F1 gain from the engineered indicators declines from 0.0528 for the Decision Tree to 0.0370 for Logistic Regression, 0.0186 for Random Forest and 0.0062 for Gradient Boosting (Holm-adjusted p = 0.106). The five indicators alone perform within 0.0030 macro-F1 of the full eleven-variable set. The findings of this study suggest that domain feature engineering and model capacity act as partial substitutes and that export vulnerability, diversion exposure and import dependency are the dominant dimensions of firm-level trade risk within the benchmark. JEL Classification Codes: C45, C53, F14, F15, M11.

American International Journal of Sciences and Engineering ResearchVol. 9(1)
Indian Institute of Information Technology Allahabad (IN)
Global trade and economics
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon