Cross-GradCAM for Point Level Explanation on Multi-Variable Time Series Data

Time series data analysis is crucial in various fields, but as data complexity increases, traditional analysis methods struggle to clarify temporal patterns. While AI models offer high accuracy, their black-box nature limits trust in critical decision-making. Explainable AI (XAI) helps by visually illustrating which variables influence predictions, improving transparency. However, existing XAI models, designed for image analysis, are inadequate for Point-level analysis in multivariate time series data. To address this, we propose Cross-GradCAM that provides Point-level explanations using a deep neural network with a specialized 1D convolution. This model precisely identifies Point-Level features at the intersection of time and variables. Experimental results highlight the efficacy of Cross-GradCAM, with the model achieving an accuracy of 99.2%, recall of 87.6%, and F1-score of 83.8%, significantly outperforming existing methods.