PredVis: Interaction Techniques for Time Series Prediction

With the increasing collection of time series data, both related to business and personal use, a substantial amount of research and development efforts are being directed to gain deeper insights from such records. Data mining techniques like similarity search and segmentation are used as tools to enhance the comprehension of this data. While innovative techniques have been examined, less work has been done in creating functional tools that use these methods for visual analysis. We have created PredVis, a mixed-initiative system that uses statistical and machine learning techniques to facilitate visual analysis of time series data for forecasting. The system provides tools to interact with model calculated results, visually query variances and integrate other dimensions of the data to improve user comprehension and decision-making. The methods developed in the system are generic and can be employed for analysis in domains including finance, health, marketing and astronomy.