Published IEEE Paper CSITSS 2025 Document ID 11295090
Lead Author Md Anisur Rahman Chowdhury

AI, cloud computing, and ERP systems research with a protected accepted-manuscript delivery workflow.

Modernizing ERP with a Hybrid AI + Cloud Framework

This portal packages the published research artifact for AI and Cloud Computing in Business Systems: A Hybrid Model for Enhancing Enterprise Resource Planning into a cleaner public-facing repository, with publication-backed visuals, architecture storytelling, and a data-conscious results dashboard.

Md Anisur Rahman Chowdhury

Gannon University, Erie, PA, USA

LinkedIn GitHub Google Scholar Portfolio ResearchGate
Open results dashboard Read HTML report Implementation guide Poster view
0% Best reported accuracy
0% F1-score of the proposed model
0 Highest module attention score
0 Training epochs reported
Overview

What the paper contributes

The paper frames ERP modernization as a combined AI and cloud problem: the model must understand module-level dependencies, while the deployment model must remain scalable and interpretable in a cloud-hosted business environment.

Problem framing

Traditional ERP systems often lack the flexibility and real-time intelligence needed for modern enterprise operations, especially when module relationships become dense and difficult to optimize.

Technical core

The proposed BiLSTM-Attention architecture captures sequential ERP dependencies from both directions and adds attention weights to expose which modules matter most to the prediction.

Enterprise relevance

The resulting predictions support risk management, intelligent load balancing, structural analysis, and cloud-aware resource decisions in ERP service ecosystems.

Publication-aligned summary

The proposed model beats the best non-proposed baseline by 5.9% accuracy points, while the most salient component in the attention analysis is Module_5 (0.17).

The site uses exact values where the paper provides them and clearly labels reconstructed elements only when raw logs are absent from the repository.

Repository cleanup goals

  • Expose the paper, source, figures, and citation artifacts in a cleaner structure.
  • Turn reported tables into machine-readable CSV and JSON files.
  • Present the work through a GitHub Pages portal, report page, poster page, and implementation guide.
Method

From ERP metadata to cloud-aware decision support

The methodology starts with ERP module dependency metadata, applies sequence-oriented preprocessing, learns contextual interactions with BiLSTM-Attention, and then maps the output to a cloud ERP integration pathway.

1. Data collection

The paper references the ERP Module-Table Dependency Dataset and models module codes, table counts, dependency counts, and complexity indicators.

2. Preprocessing

Missing values are handled, categorical data is encoded, numerical features are normalized, and linked modules are converted into ordered sequences.

3. Hybrid modeling

BiLSTM captures forward and backward sequence context, while attention weights highlight the most influential ERP modules.

Methodology workflow diagram showing data collection, preprocessing, AI model training, cloud integration, and evaluation.
Methodology figure converted from the paper source bundle.
Diagram of the BiLSTM-Attention architecture for ERP module dependency prediction.
Published architecture diagram for the proposed BiLSTM-Attention model.

Architecture explorer

Data collection and ERP metadata modeling

    Cloud-native ERP architecture with a BiLSTM-Attention microservice and monitoring components.
    Cloud-native integration diagram included with the paper source bundle.

    What is verified vs conceptual

    • Verified in the paper: model design, reported metrics, attention table, hyperparameters, and training behavior figures.
    • Conceptual but paper-backed: the cloud-native ERP integration pathway for autoscaling, fault tolerance, and microservice deployment.
    • Not present in the repository: raw ERP dataset, live deployment code, and original per-epoch training logs.
    Results

    Interactive results dashboard

    These views combine exact reported metrics with charted comparisons and one clearly identified reconstructed training series derived from the published figures.

    89.5% Precision
    90.8% Recall
    90.1% F1-score
    85.3% Best baseline accuracy (Random Forest)

    Exact baseline comparison

    All bar values come directly from the model comparison table reported in the paper.

    Attention allocation by module

    Exact attention shares are taken from the module interpretation table. They sum to 1.00.

    Model-shape comparison

    The proposed model improves across all four reported evaluation dimensions, not just a single metric.

    Reconstructed training behavior

    This line chart is reconstructed from the published training figures because raw epoch logs were not included in the repository.

    Visual Lab

    Six supporting analytics views

    The README and website both surface the required visualization set. Every chart here is grounded in the paper's reported tables, attention scores, or clearly marked reconstructed training traces.

    DistPlot

    Distribution plot of ERP module attention scores.

    Distribution of the 10 exact attention scores reported for ERP modules.

    Pie Chart

    Pie chart of attention share across ERP modules.

    Share view of the same exact module attention weights, highlighting Module_5 as the dominant component.

    ViolinPlot

    Violin plot showing reported metrics grouped by model family.

    Small-sample violin view of reported metric distributions across traditional ML, sequence DL, and proposed model families.

    HeatMap

    Heatmap of baseline model metrics.

    Matrix rendering of the exact model comparison table from the paper.

    PairPlot

    Pairplot of model precision, recall, F1-score, and accuracy.

    Scatter-matrix view of the four evaluation metrics across the five published models.

    JointPlot

    Joint plot of precision and recall values across models.

    Precision versus recall relationship with marginal histograms, built from exact table values.

    Cloud ERP

    Business-system implications of the model

    The paper does not claim a production deployment, but it formalizes how attention-aware dependency prediction can support cloud ERP operations in realistic service environments.

    Intelligent load management

    High-dependency modules can be identified earlier and assigned more compute or routing priority in cloud-hosted ERP systems where module demand varies sharply.

    Fault tolerance and risk management

    Modules with consistently high attention can be monitored as possible dependency hubs or failure multipliers inside complex ERP ecosystems.

    Billing and modular transparency

    The paper suggests that dependency-aware compute weighting could support more transparent cost attribution for modules with heavier computational paths.

    Dynamic resource allocation

    R_i = alpha_i * C_i

    Compute allocation is scaled by the attention-informed importance of each module.

    Dependency centrality

    D_k = sum I(M_k in T_i) * alpha_ik

    The paper uses this formalism to discuss fault-prone central dependencies in ERP workflows.

    Inference as a cloud microservice

    f_theta: T -> c -> y_hat

    The BiLSTM-Attention engine is framed as a stateless service that can interface with modular ERP services.

    Publication integrity note

    The model is experimentally validated for prediction quality. The operational cloud use cases remain a paper-backed integration pathway, not a claimed deployed production system.

    Accepted Manuscript

    Protected PDF and LaTeX source in the same style as the follow-up project

    The accepted manuscript package for this ERP research portal follows the same public-access pattern as the follow-up site: profile confirmation, video/channel confirmation, author email request, and encrypted archive delivery.

    IEEE CSITSS 2025 Accepted Manuscript

    AI and Cloud Computing in Business Systems: A Hybrid Model for Enhancing Enterprise Resource Planning

    Md Anisur Rahman Chowdhury, Khandakar Rabbi Ahmed, Kefei Wang, Sabrina Mohona, Shahriar Alam Robin, and Shah Tawkir Nesar

    Gannon University, International University of Business Agriculture and Technology, and Lamar University

    Abstract

    The paper combines AI and cloud computing to improve ERP adaptability, predictive insight, and scalability. A BiLSTM-Attention model is used to learn ERP module relationships and support modular analysis in cloud-hosted enterprise systems.

    The reported results reach 91.2% accuracy, 89.5% precision, 90.8% recall, and 90.1% F1-score, with Module_5 receiving the highest attention score at 0.17.

    Highlights

    The accepted manuscript section is now protected with encrypted archive downloads, while the public portal continues to expose the figures, metrics, HTML report, poster, and implementation notes.

    PDF and LaTeX source packages are available through the secure gate below, matching the follow-up manuscript-access workflow.

    Video

    Research overview walkthrough

    The repository also links to the author's public video overview, giving visitors a second way to understand the motivation, architecture, and outcomes of the work.

    Why include the video here

    The site already exposes the paper, source files, and charted results. The embedded overview adds a communication layer for readers who want a faster narrative path through the same published work.

    Open on YouTube Read the HTML report
    Assets

    What the repository now exposes

    The project has been reorganized so the paper, source files, chart assets, and supporting pages are easier to discover and reuse.

    Accepted manuscript package

    The public website now exposes the accepted manuscript through encrypted archives only, matching the protected access flow used in the follow-up project.

    IEEE record

    Machine-readable research data

    Exact metrics, attention scores, and reconstructed training curves are published as CSV and JSON for traceable reuse.

    research_data.json Data folder

    Alternative views

    Beyond the main portal, the site now includes a narrative HTML report, a deployment-aware implementation guide, and a poster-style summary page.

    Report Implementation Poster
    Publication metadata note: Conference and publication metadata are aligned to the official indexing record. The bundled LaTeX source retains an older first-page header from an earlier template.
    Author Profiles

    Md Anisur Rahman Chowdhury across all public research profiles

    This site now highlights the lead author and collects all public profiles in one section for verification, citation, collaboration, and manuscript-password requests.

    Portfolio

    Personal site and portfolio showcase.

    marcbd.site

    Contact

    Password requests and research inquiries can be sent directly by email.

    engr.aanis@gmail.com
    Citation

    How to cite and explore the research

    Citation files are available in both BibTeX and GitHub-native formats, and the repository now makes the key publication paths obvious.

    BibTeX

    @inproceedings{chowdhury2025erpcloudai,
  author    = {Md Anisur Rahman Chowdhury and Khandakar Rabbi Ahmed and Kefei Wang and Sabrina Mohona and Shahriar Alam Robin and Shah Tawkir Nesar},
  title     = {AI and Cloud Computing in Business Systems: A Hybrid Model for Enhancing Enterprise Resource Planning},
  booktitle = {2025 9th International Conference on Computational System and Information Technology for Sustainable Solutions (CSITSS)},
  pages     = {1--6},
  year      = {2025},
  url       = {https://ieeexplore.ieee.org/abstract/document/11295090},
  note      = {IEEE Xplore document 11295090}
}

    Repository citation metadata is also published via `CITATION.cff`.