Biased vs. Unbiased Data in Machine Learning for Predicting Properties of Polymer Nanocomposites

EasyChair Preprint 14501

Abstract

Machine learning (ML) has emerged as a powerful tool for predicting the properties of polymer nanocomposites, offering significant advantages in material design and optimization. However, the reliability of these predictions is heavily influenced by the nature of the data used in model training. This paper explores the critical distinction between biased and unbiased data in the context of ML-driven predictions for polymer nanocomposites. Biased data, often resulting from imbalanced datasets or systematic errors in experimental procedures, can lead to skewed model outputs that fail to generalize across diverse material systems. Conversely, unbiased data, characterized by balanced and representative sampling, enhances the accuracy and robustness of predictive models. We analyze the implications of data bias on model performance, including the potential for overfitting and the propagation of inaccuracies in property predictions. Furthermore, strategies for mitigating data bias, such as advanced data augmentation techniques and the integration of domain knowledge, are discussed. The findings underscore the necessity of rigorous data management practices to ensure the development of reliable and generalizable ML models in the field of polymer nanocomposites

Keyphrases: Machine Learning Models, Unbiased Data, polymer nanocomposites

@booklet{EasyChair:14501,
  author    = {Abi Cit},
  title     = {Biased vs. Unbiased Data in Machine Learning for Predicting Properties of Polymer Nanocomposites},
  howpublished = {EasyChair Preprint 14501},
  year      = {EasyChair, 2024}}