Human Visual Attention data contains important information about how people interact with their environment, which can be used to optimize information visualizations for effectiveness and to build adaptive, interactive systems. In this thesis, the relationship and characteristic patterns between visual analytical tasks on visualizations and task-driven visual attention will be analyzed, while research so far has mainly focused on task-agnostic attention. In particular, by predicting the question type based on task-driven attention data on visualizations, this work contributes to a better understanding of attention patterns and visual behavior in analytical tasks. To this end, ChartQC is presented, a multi-modal neural network that integrates attention data with spatial context information from the chart images and includes statistical saliency metrics, creating a basis for informed question type classification. In an experiment, ChartQC outperformed other baseline models in direct comparison across several important evaluation metrics, but also achieved better results compared to variations of the ChartQC architecture in the context of an ablation study. This work demonstrates the feasibility of extracting contextual information from task-driven visual attention data. ChartQC’s architecture, combined with the findings from the qualitative analysis and the discussion, offers a solid foundation for future research on task-driven attention and adaptive systems.