A Selective State-Space Model for Real-Time E-Commerce Transaction Anomaly Detection in Management Information Systems

Management information systems (MISs) are subject to complex patterns of fraud, which change over time, as they occur at high speeds in class-imbalanced data streams. E-commerce platforms process very large transaction volumes. Existing detectors have three structural problems. The quadratic complexity of transformer attention makes it difficult to meet millisecond-scale inference budgets, recurrent models suffer from gradient pathologies over long behavioral sequences, and static loss functions are unable to adjust to the severe and changing class imbalance found in real transaction logs. This article introduces a selective state-space framework (SHOPMAMBA-MIS) that tackles all three challenges within a single framework. SHOPMAMBA-MIS consists of three parts: Selective-State Autoencoder (SSAE) to compress behavioral transaction sequences with linear-time selectivity, Transaction Attention Module (TAM) to emphasize anomaly boundaries per token without complete self-attention, and Dynamic Imbalance-Focal Sampler (DIFS) to dynamically sample the focal loss weight per mini-batch based on an exponential moving average (EMA) class-ratio estimator. On three public benchmarks, IEEE-CIS Fraud Detection, PaySim, and ULB Credit Card Fraud, SHOPMAMBA-MIS outperforms the best transformer baseline by +4.2 percentage points (F1), +5.9 percentage points (F1), and +4.1 percentage points (F1), respectively, while on the same benchmarks, its inference latency is reduced to 1.8 ms per transaction batch, which is 6.3 times faster than the slowest transformer baseline, TranAD (11.3 ms per batch), and its detection recall is kept above 92.8% (over 97.9% on IECIS and ULB) in all three benchmarks at a false-positive rate below 0.5%. The results obtained demonstrate the feasibility of the SHOPMAMBA-MIS, which can be implemented as an intelligent component in enterprise Management Information Systems, for real-time monitoring, enterprise control, and managerial decision-making in an enterprise. The framework is specifically suited to payment authorization pipelines that provide behavioral features (including rolling account-level aggregates where identifiers exist) and at least L prior transactions of history, exhibit intra-day fraud-rate variation, and enforce synchronous millisecond-scale latency constraints; applicability conditions and boundary cases.

Authors

Institutions

Publication Details

Journal
Journal of theoretical and applied electronic commerce research
Published
2026-11-19
DOI
https://doi.org/10.3390/jtaer21110430
Primary Topic
Imbalanced Data Classification Techniques
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

A Selective State-Space Model for Real-Time E-Commerce Transaction Anomaly Detection in Management Information Systems

Ziad Hmwd A Almtiri
Journal of theoretical and applied electronic commerce research
Imbalanced Data Classification Techniques
article

A Selective State-Space Model for Real-Time E-Commerce Transaction Anomaly Detection in Management Information Systems

Ziad Hmwd A Almtiri
article en

Abstract

Management information systems (MISs) are subject to complex patterns of fraud, which change over time, as they occur at high speeds in class-imbalanced data streams. E-commerce platforms process very large transaction volumes. Existing detectors have three structural problems. The quadratic complexity of transformer attention makes it difficult to meet millisecond-scale inference budgets, recurrent models suffer from gradient pathologies over long behavioral sequences, and static loss functions are unable to adjust to the severe and changing class imbalance found in real transaction logs. This article introduces a selective state-space framework (SHOPMAMBA-MIS) that tackles all three challenges within a single framework. SHOPMAMBA-MIS consists of three parts: Selective-State Autoencoder (SSAE) to compress behavioral transaction sequences with linear-time selectivity, Transaction Attention Module (TAM) to emphasize anomaly boundaries per token without complete self-attention, and Dynamic Imbalance-Focal Sampler (DIFS) to dynamically sample the focal loss weight per mini-batch based on an exponential moving average (EMA) class-ratio estimator. On three public benchmarks, IEEE-CIS Fraud Detection, PaySim, and ULB Credit Card Fraud, SHOPMAMBA-MIS outperforms the best transformer baseline by +4.2 percentage points (F1), +5.9 percentage points (F1), and +4.1 percentage points (F1), respectively, while on the same benchmarks, its inference latency is reduced to 1.8 ms per transaction batch, which is 6.3 times faster than the slowest transformer baseline, TranAD (11.3 ms per batch), and its detection recall is kept above 92.8% (over 97.9% on IECIS and ULB) in all three benchmarks at a false-positive rate below 0.5%. The results obtained demonstrate the feasibility of the SHOPMAMBA-MIS, which can be implemented as an intelligent component in enterprise Management Information Systems, for real-time monitoring, enterprise control, and managerial decision-making in an enterprise. The framework is specifically suited to payment authorization pipelines that provide behavioral features (including rolling account-level aggregates where identifiers exist) and at least L prior transactions of history, exhibit intra-day fraud-rate variation, and enforce synchronous millisecond-scale latency constraints; applicability conditions and boundary cases.

Journal of theoretical and applied electronic commerce researchVol. 21(11)
Taif University (SA)
Imbalanced Data Classification Techniques
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