Smooth Operator: Step Towards a Generalized Smoothing Framework

Report : Smooth Operator: Step Towards a Generalized Smoothing Framework

Today’s visualization systems typically plot noisy raw data that obscures long-term trends from the user. In this project, we suggested steps toward developing a general framework to automate time-series smoothing. We explored different smoothing functions and statistical measures for smoothing a time series. We analyzed the results of our user study qualitatively and quantitatively to explore correlations between different time series and optimum smoothing techniques.

• The figure above represents the variation in smoothing of the weekly homicides in Chicago dataset by changing the smoothing function and statistical measure.

• We use a combination of statistical measure and smoothing function to smooth a time series. Further, we use the standard deviation of the first differences of the smoothed time series as a measure of roughness of the time series. For a given combination of statistical measure and smoothing function, our algorithm computes the optimal smoothing length.

• We conducted a user study to gauge people’s preferences of the best combinations of smoothing function and statistical measure for different datasets.

• Our recommendations for dealing with seasonal and noisy time series are based on extensive evaluations stated in our Paper, and we believe that they can be helpful in improv- ing the reliability and accuracy of smoothing time series.