The use of kernel densities and confidence intervals to cope with insufficient data in validation experiments

A novel procedure to establish probability density distributions based on insufficient data is introduced. The approach requires the selection of a confidence level to cover the unknown distribution. The use of kernel densities is proposed for its high fidelity to data and its capability to represent correlations correctly. It is demonstrated that the proposed approach can be applied successfully for validation, leading to measures for the validity in the validation and accreditation experiments and provides consistent predictions based on the required confidence level and the size of available data points of the calibration data.