Assessing Data Reliability for AI-Driven Volcanic Rock Dating: A Comparison of Electron Microprobe and Laser Ablation Mass Spectroscopy

Abstract

This study explores the integrationof artificial intelligence (AI) and modern data analytics for accurately predicting and classifying three distinct periods of volcanic activity. By leveraging previously dated volcanic samples, we assess whether existing age and geochemical data can reliably group and predict volcanic episodes. Our study focuses on the Kula Volcanic Province (Turkey). We compare the effectiveness of two analytical techniques—Electron Microprobe Analysis (EPMA) and Laser Ablation Inductively Coupled Plasma Mass Spectrometry (LA-ICP-MS)—in producing high-quality datasets for training deep learning models. While EPMA provides major and minor elemental compositions, LA-ICP-MS offers a broader range of trace elements, which may improve classification accuracy. Two experiments were conducted to evaluate the feasibility of AI-based volcanic rock age estimation. In the first experiment, an autoencoder and unsupervised clustering were applied to reduce dimensionality and group samples based on their elemental composition. The results revealed that EPMA data lacked sufficient detail to form well-defined clusters, whereas LA-ICP-MS data produced clusters that closely aligned with true age classes due to their higher sensitivity to trace elements. In the second experiment, a deep neural network (DNN) was trained to classify rock ages. The LA-ICP-MS-based model achieved a classification accuracy of 95 %, significantly outperforming the EPMA-based model (72 %). These findings underscore the importance of data quality and analytical technique selection in AI-powered geochronology, demonstrating that high-quality trace element data enhances AI model performance for volcanic rock age estimation.