Automated machine learning for fabric quality prediction: a comparative analysis

Ahmet Metin ¹, Turgay Tugay Bilgin ¹

⁰Bursa Technical University, Bursa, Turkey.

Peerj. Computer Science +

|

August 15, 2024

View abstract on PubMed

Summary

This study evaluates seven automated machine learning (AutoML) tools for textile fabric quality prediction using Industry 4.0 data. EvalML and AutoGluon showed superior performance in different accuracy metrics, balancing efficiency and forecast accuracy.

Area Of Science

Textile Manufacturing
Machine Learning
Industry 4.0

Background

Fabric quality prediction is crucial in textile manufacturing.
Internet of Things (IoT) and Enterprise Resource Planning (ERP) systems offer valuable data.
Industry 4.0 integration enhances productivity and reduces lead times.

Purpose Of The Study

To address imbalanced data challenges in fabric quality prediction.
To evaluate seven open-source automated machine learning (AutoML) technologies.
To identify optimal AutoML solutions balancing computational efficiency and forecast accuracy.

Main Methods

Evaluation of seven AutoML tools: FLAML, AutoViML, EvalML, AutoGluon, H2OAutoML, PyCaret, and TPOT.
Development of an innovative approach to compromise between computational efficiency and forecast accuracy.
Analysis of feature importance rankings for model interpretability.

Main Results

EvalML performed best for a specific objective function, excelling in Mean Absolute Error (MAE).
AutoGluon demonstrated superior performance in Mean Absolute Percentage Error (MAPE), Root Mean Squared Error (RMSE), and R-squared, despite longer inference times.
Sin/cos encoding proved effective for categorical variables with numerous unique values.

Conclusions

AutoML technologies offer significant potential for enhancing fabric quality prediction in the textile industry.
Balancing predictive accuracy and computational efficiency is key for practical AutoML implementation.
Feature importance analysis enhances model interpretability, guiding future research and Industry 4.0 adoption.

Keywords:

AutoML Feature importance Hyperparameter optimization Imbalanced data Model interpretability Quality control Textile industry