Empowering Capacitive Devices: Harnessing Transfer Learning for Enhanced Data-Driven Optimization.

Area of Science:

• Engineering
• Machine Learning
• Data Science

Background:

• Data-driven models are crucial for engineering tasks like material design and process monitoring.
• Capacitive devices (deionization, supercapacitors) could benefit from machine learning (ML) for energy-efficient applications.
• Limited and incomplete datasets hinder the development of effective data-driven ML models.

Purpose of the Study:

• To address challenges posed by limited and incomplete datasets in developing data-driven models.
• To explore the efficacy of transfer learning in conjunction with data imputation for model enhancement.
• To develop a novel two-step ML modeling framework, ImputeNet, for improved predictive capabilities.

Main Methods:

• Developed ImputeNet, a two-step framework involving ML-imputed datasets and subsequent training on clean datasets.
• Utilized transfer learning to overcome limitations associated with sparse or missing data.
• Applied genetic algorithms for optimization studies, analyzing solutions under Pareto optimality.

Main Results:

• Demonstrated the capability to develop data-driven models with acceptable metrics that mirror experimental measurements.
• Successfully employed data imputation and transfer learning to enhance model performance with incomplete data.
• Provided early insights for identifying promising experimental conditions, accelerating research and development.

Conclusions:

• ImputeNet framework effectively handles incomplete datasets, enabling robust data-driven modeling.
• Transfer learning and data imputation are viable strategies for improving ML model performance in data-scarce environments.
• The approach offers potential for accurate predictive modeling across diverse fields, including healthcare and environmental monitoring.