README – GA-FL-Transformer: Intelligent Optimization Framework for Corporate Financial Management

Title

GA-FL-Transformer: An Intelligent Optimization Framework for Corporate
Financial Management

Description

This repository contains the implementation code and dataset
preprocessing scripts for the research paper titled “Design of an
Intelligent Optimization Framework for Corporate Financial Management
Based on GA-FL-Transformer.” The GA-FL-Transformer is a hybrid
optimization model that integrates Genetic Algorithm (GA), Fuzzy Logic
(FL), and Transformer architecture to address challenges in enterprise
financial management. It aims to enhance financial decision-making by
achieving multi-source data perception, adaptive optimization, and
interpretable reasoning in dynamic business environments. The framework
performs unified encoding of heterogeneous financial data,
attention-guided genetic-fuzzy optimization, and interpretable
rule-based reasoning, providing a closed-loop decision-making process
from perception to optimization.

Dataset Information

Two publicly available financial datasets were employed in this
study: 1. Compustat Dataset - Source: Zenodo Repository
(https://zenodo.org/records/15369177) - DOI: 10.5281/zenodo.15369177 -
Description: Contains structured financial statements (balance sheet,
income statement, cash flow) for North American listed firms. 2. CRSP
Dataset (CRISPR Therapeutics AG Stock Performance) - Source: Zenodo
Repository (https://zenodo.org/records/12555376) - DOI:
10.5281/zenodo.12555376 - Description: Provides market behavior data
such as stock prices, returns, and trading volumes for financial
performance and risk analysis.

Preprocessing steps: - Normalization of numerical features (z-score) -
TF-IDF and Word2Vec embedding for textual data - Sliding-window
smoothing for temporal sequences - Gaussian imputation for missing
values - Unified embedding concatenation for Transformer input (see Eq.
(2) in the paper)

Code Information

The repository contains the following main modules: -
data_preprocessing.py — Prepares and encodes multi-source data
(structured, text, time series). - transformer_encoder.py — Implements
multi-head self-attention for feature perception. -
genetic_fuzzy_module.py — Performs genetic optimization and fuzzy rule
adaptation. - model_training.py — Handles joint training, validation,
and optimization of GA-FL-Transformer. - evaluation_metrics.py —
Calculates Precision, Recall, F1-Score, and Error Rate. -
ablation_analysis.py — Conducts ablation experiments for model component
verification. - config.yaml — Contains hyperparameter settings (learning
rate, batch size, epochs, crossover/mutation rates).

Usage Instructions

1.  Installation

    git clone https://github.com/<your-org>/GA-FL-Transformer.git
    cd GA-FL-Transformer
    pip install -r requirements.txt

2.  Dataset Preparation Download datasets from the Zenodo links above
    and place them in the ./data/ directory:

    data/
     ├── compustat.csv
     ├── crsp.csv

3.  Model Training

    python model_training.py --dataset compustat --epochs 200 --batch_size 64

4.  Evaluation and Visualization

    python evaluation_metrics.py --dataset crsp
    python ablation_analysis.py

Requirements

-   OS: Windows 10/11 or Ubuntu 22.04
-   Python ≥ 3.10
-   Libraries: torch >= 2.2.0 numpy >= 1.25.0 pandas >= 2.1.0
    scikit-learn >= 1.3.0 gensim >= 4.3.0 matplotlib >= 3.7.0 tqdm >=
    4.66.0
-   GPU: CUDA 12.1 + cuDNN 8.9 (optional)

Methodology

The GA-FL-Transformer framework comprises three core layers: 1.
Perception Layer: Transformer encoder extracts multi-source financial
features via multi-head attention. 2. Optimization Layer: Genetic
Algorithm evolves fuzzy rules and membership functions guided by
attention weights. 3. Decision Layer: Fuzzy Logic performs interpretable
reasoning and outputs financial decisions. The training process
alternates between Transformer feature updates and GA-FL optimization to
achieve closed-loop convergence.

Citations

If you use this code or dataset in your research, please cite: Zhu F.,
Liu S., Yuan F., & Arshad M. (2025). Design of an Intelligent
Optimization Framework for Corporate Financial Management Based on
GA-FL-Transformer. Journal of Computational Methods in Sciences and
Engineering.

License & Contribution Guidelines

-   License: MIT License – free for academic and research use with
    proper attribution.
-   Contributions: Contributions, pull requests, and extensions (e.g.,
    alternative optimizers or Transformer variants) are welcome. Please
    open an issue before submitting major changes.
-   Contact: For questions, email jenny2025226@163.com (corresponding
    author).