Time series, as we call sequences of measurements of an observed phenomenon, represent an important type of data in the fields of econometrics (e.g. countries' yearly GDP and relative debt change), business (e.g. number of products sold per month), medicine (EEG, ECG), meteorology (e.g. change in average temperature through time) and in almost all other fields of natural and social science. It is thus important for toolsets to exist, with which one can analyze, transform, visualize, and model time series data. Based on renowned Orange data mining software framework, we propose a suite of visual programming widgets for construction of workflows for interactive time series analysis, visualization, and forecast. In particular, the suite comprises widgets for time series differencing, interpolation, aggregation, seasonal adjustment, transformation with window functions and estimation of causality. Additionally, we devise components for plotting time series data in a line chart diagram, periodogram, correlogram, and spiral heatmap. We support time series modeling with VAR or ARIMA models. We evaluate our contribution on various time series data sets.
|
