Investigating the Influence of Biased Data on Predictive Modeling of Polymer Nanocomposites Using Artificial Intelligence and Machine Learning

EasyChair Preprint 14514

Abstract

The integration of artificial intelligence (AI) and machine learning (ML) in materials science has revolutionized the predictive modeling of polymer nanocomposites, enabling the rapid discovery and optimization of novel materials with superior properties. However, the quality and bias of the input data play a critical role in determining the accuracy and generalizability of these predictive models. This study investigates the influence of biased data on AI-driven predictive modeling of polymer nanocomposites, focusing on how skewed or incomplete datasets can lead to erroneous predictions and suboptimal material designs. We analyze the impact of various forms of bias, including sampling bias, measurement error, and feature selection bias, on the performance of ML models in predicting key mechanical, thermal, and electrical properties of polymer nanocomposites. Through a series of computational experiments, we demonstrate how biased data can distort the relationship between input features and material properties, leading to models that fail to generalize across different material systems or environmental conditions. Additionally, we explore strategies for mitigating the effects of biased data, such as data augmentation, synthetic data generation, and the incorporation of domain knowledge into the modeling process. The findings of this research underscore the importance of data quality and integrity in AI-driven materials design, offering insights for developing more robust and reliable predictive models for polymer nanocomposites.

Keyphrases: Artificial Intelligence, polymer nanocomposites, predictive modeling

@booklet{EasyChair:14514,
  author    = {Abey Litty},
  title     = {Investigating the Influence of Biased Data on Predictive Modeling of Polymer Nanocomposites Using Artificial Intelligence and Machine Learning},
  howpublished = {EasyChair Preprint 14514},
  year      = {EasyChair, 2024}}