Year: 2026
Type: article
Abstract: Automation, artificial intelligence, connected sensing technologies, digital twins, robotics, and data-driven control systems are increasingly transforming textile and apparel production systems. However, in existing studies, these technologies have often been analyzed separately or limited to specific factories, countries, or specific stages of the production process. The study created an integrated analytical framework for analyzing the maturity of technology adoption in modern textile production systems, with a new dataset based on data collected from various sources related to technology adoption and digital readiness open to the public. The complete dataset contains a total of 8,198 analytical observations and 26 variables. The dataset covers nine manufacturing and business activities for the period 2019-2023 in five countries - Bangladesh, Chile, Ethiopia, India and Senegal. As part of this study, a subset of 5,472 production-based observations were used for predictive modeling. This subset is divided into 3,830 trainings, 821 validations, and 821 test observations, respectively, in a group-conscious manner. The study compared the performance of four models: Random Forest, XGBoost, Support Vector Machine and One-Dimensional Convolutional Neural Network (1D CNN). The final results showed 92.00%, 93.00%, 93.56% and 96.10% accuracy respectively, making 1D CNN the highest performing model. In a test set of 821 observations, 1D CNN's class-based F1 score was 95.29% for Basic / Manual, 97.54% for Intermediate / Mechanized, 68.57% for Digital, and 95.20% for Advanced / Smart. In the supplied ROC analysis, the value of AUC was found to be 0.998094. Integrated Gradients-based Explainable AI analysis also indicates that country-specific context, clothing cutting and sewing activities, production-domain indicators, quality-management-related variables, survey duration, and internet access rate contribute significantly to model decisions. Overall, the results show that it is possible to effectively analyze the maturity of technology adoption by combining various publicly available technology indicators with machine learning and deep learning. At the same time, class inequalities and limitations in the comparability of data from different countries have also been identified, which are important considerations for future research and external verification.
Institution: University of Dhaka
Language: English
Subfield: Industrial and Manufacturing Engineering
Field: Engineering
Domain: Physical Sciences
Open Access status: hybrid
OpenAlex ID: W7223204357