AutoFuzzTS

Time series forecasting library using fuzzy logic and automated machine learning.
Build and evaluate time series models automatically using fuzzy logic and AutoML techniques.

The package is designed for academic benchmarking and controlled experimentation in one-step-ahead time-series forecasting. It assumes a fixed train/validation/test split and focuses on reproducible model comparison rather than real-time deployment.

Installation

pip install autofuzzts

🚀 Quick Start

Load and prepare your time series data

import pandas as pd

# Load dataset into a pandas DataFrame
data = pd.read_csv('../../data/sample_datasets/NYSE.csv')
data.head(10)

# Select the target column to forecast
data_column_name = "Close"
df = data[[data_column_name]].copy()

# Split into train, validation, and test sets
test_len = len(df) // 5
val_len = len(df) // 5
train_len = len(df) - test_len - val_len

df_train = df[:train_len]
df_val = df[train_len:(train_len + val_len)]
df_test = df[(train_len + val_len):]

Tune hyperparameters using Bayesian search

from autofuzzts import pipeline

# Run Bayesian optimization for fuzzy pipeline configuration
best_config, best_rmse = pipeline.tune_hyperparameters_bayes(
    train_set=df_train,
    val_set=df_val,
    n_trials=20,
    metric="rmse"
)

print(f"Best configuration: {best_config}")

Example output:

Best configuration: {'n_fuzzy_sets': 13, 'number_of_lags': 6, 'fuzzy_part_func': 'Cosine'}

Train, calibrate, and predict

from autofuzzts import fit_calibrate_predict

# Train model, calibrate, and make one-step-ahead predictions
pred_set, pred_center_points, pred_test = fit_calibrate_predict(
    train_set=df_train,
    test_set=df_test,
    config=best_config,
    model_type="xgb"
)

This returns:

• pred_set: predicted fuzzy sets
• pred_center_points: corresponding fuzzy center values
• pred_test: crisp numeric predictions (one-step-ahead forecast)

Function Overview

fit_calibrate_predict()

fit_calibrate_predict(
    train_set: pd.DataFrame,
    test_set: pd.DataFrame,
    config: dict,
    model_type: Literal['xgb', 'mlp', 'tpot'] = 'xgb',
    number_cv_calib: int = 5,
    diff_type: Literal['perc', 'abs'] = 'perc',
    covariates: list[str] | None = None,
    exclude_bool: bool = False
) -> float

Trains and calibrates a fuzzy time series model on the training set using cross-validation, then predicts on the test set and returns performance metrics.

Description

AutoFuzzTS automates the process of fuzzy time series modeling by:

• building and testing multiple fuzzy pipelines,
• tuning hyperparameters using Bayesian optimization, and
• integrating tuned classification models - XGBoost, MLP, or TPOT.

This allows for rapid experimentation and selection of optimal configurations for forecasting tasks.

📄 Reference

This code is based on the research:

Optimizing stock price forecasting: a hybrid approach using fuzziness and automated machine learning
Jan Timko, Radwa El Shawi, Stefania Tomasiello
Expert Systems with Applications, Volume 259, 2025, 128844

Read on ScienceDirect

If you use this code in your research or projects, please cite the paper.

📄 License

This project is licensed under the MIT License.