financial-model-taiwan 1.0.3

A financial model package trained on data from Taiwan fiscal year 1999-2009

Financial Model Taiwan

Financial Model Taiwan is a Python package designed to preprocess, train, and predict financial models. It includes functionalities for data ingestion, preprocessing, resampling, model training, evaluation, and optimization. This package aims to provide a comprehensive solution for financial modeling with support for various machine learning algorithms and techniques.

Model Architecture

The following image illustrates the model architecture:

Workflow Overview

1. Data Ingestion:

   • Load data from CSV files.
   • Split data into training and testing sets.

2. Preprocessing Pipeline:

   • Define preprocessing steps.
   • Handle missing values.
   • Standardize/normalize data.
   • Select important features.

3. Resampling:

   • Perform data resampling to handle class imbalance.

4. Initial Model Training:

   • Train multiple models: Random Forest, Logistic Regression, SVM, Gradient Boosting, XGBoost, KSBBoost, ANN.
   • Evaluate models based on performance metrics.
   • Select the best performing models for stacking.

5. Optimizing Model for Performance:

   • Stack the best models (XGBoost and Random Forest).
   • Optimize hyperparameters using Optuna.
   • Tune the stacked model.
   • Adjust thresholds to minimize type I and type II errors.
   • Increase recall and finalize the model.

Installation

You can install the package via pip:

pip install financial_model_taiwan

Usage

Training a New Model

from financial_model_taiwan.Orchestrator import FinModel

pipeline = FinModel(data_path='data/train_data.csv', target_column='target')
pipeline.data_ingestion()
pipeline.data_preprocessing()
pipeline.data_resampling()
pipeline.train_model()
pipeline.save_model('models/trained_model.bin')
evaluation_results = pipeline.evaluate_model()
print(evaluation_results)

Using a Pre-trained Model

from financial_model_taiwan.Orchestrator import FinModel

pipeline = FinModel(data_path='data/train_data.csv', target_column='target', model_path='models/trained_model.bin')
pipeline.data_ingestion()
pipeline.data_preprocessing()
pipeline.load_model()
evaluation_results = pipeline.evaluate_model()
print(evaluation_results)

FinModel Class Attributes

• data_path: Path to the CSV file containing the dataset.
• target_column: The target column name for prediction.
• model_path: Path to the pre-trained model file (optional).
• pipeline: The preprocessing pipeline.
• trainer: The model trainer.
• X_train: Training features.
• X_test: Testing features.
• y_train: Training labels.
• y_test: Testing labels.
• X_train_transformed: Transformed training features after preprocessing.
• X_test_transformed: Transformed testing features after preprocessing.
• X_resampled: Resampled training features to handle class imbalance.
• y_resampled: Resampled training labels to handle class imbalance.

Example Workflow

from financial_model_taiwan.Orchestrator import FinModel

pipeline = FinModel(data_path='../data/data.csv', model_path='../models/trained_model.bin', target_column='Bankrupt?')
pipeline.data_ingestion()
pipeline.data_preprocessing()
pipeline.data_resampling()
pipeline.train_model()
pipeline.save_model('trained_model_new.bin')
pipeline.load_model()
evaluation_results = pipeline.evaluate_model()
print(evaluation_results)

# Accessing Attributes
print(pipeline.X)
print(pipeline.y)
print(pipeline.X_test_transformed)
print(pipeline.X_train_transformed)
print(pipeline.X_resampled)
print(pipeline.y_resampled)

To check the predicted values

Although we didn't explicitly create a method we can call it using

pipeline.trainer.stacked_model.predict(pipeline.X_test_transformed)