A Pattern Adaptive Technique to Handle Data Quality Variation

摘要

This study proposes a new technique, namely the Pattern Adaptive Neural Network (PANN), for simplifying existing noise detection and removal methods. This technique is developed based on a modified backpropagation algorithm using a fuzzy membership function on the error term. It is able to make use of noisy data in a single step, with an automatic adjustment of data contribution to network training. It is demonstrated via an application on an oil well data set. The results show that the predictions from PANN matched well with the expert interpretations on the data set regarding the data quality.