Comparison of Data Interpolation Methods in Time Course Pupil Diameter Data

The missing data problem is one of the main challenges in many datasets. As long as the percentage of loss is under an acceptable range, different methods can be performed in order to fill these unobserved values. In this study the thresholding method, polynomial regression approach, smoothing splines, piecewise linear interpolation and the moving median approaches are used in order to fill the missing data. Among these alternatives, the smoothing spline method typically gives higher accuracy and captures the global feature of the data, whereas, it can eliminate the local changes in the measurements while smoothing. Hereby, in this study, we propose some alternative approaches, called normal ratio and normal ratio weighted with correlation together with modified moving median method in order to fill the missing data. These novel methods are previously applied in meteorological studies where the location of the missing values in a time-course dataset is important.