Towards Enhancing Data Imputation Techniques Through Graph Attention Autoencoder Models

<p dir="ltr">The exponential growth of digital data presents significant challenges in maintaining data quality, particularly regarding missing values. Incomplete or low-quality data can lead to biased analyses and poor decision-making, undermining the effectiveness of data-driven systems. While existing imputation methods address missing data, they often struggle with complex datasets where capturing intricate relationships is necessary. A novel Autoencoder-based Graph Neural Network is proposed to address this limitation, leveraging graph structures to model relationships among data points and enhance imputation accuracy beyond traditional methods. Our model integrates graph attention mechanisms within an autoencoder framework to focus on essential features and relationships, effectively imputing missing values. By employing tailored strategies, the model adeptly handles both numerical and categorical data, ensuring versatility across various datasets. Experimental results indicate that our approach outperforms current methods by achieving 4% higher imputation accuracy. This work contributes to a more accurate and scalable solution for missing data imputation, advancing the field of data quality management.</p>