It is normal to find at least a few measured values in CO2-alkanolamine-H2O datasets that deviate greatly from the majority of published data, as the data come from different sources. These values, termed as data outliers, are the major source of conflict in modeling, simulation and process development studies. Therefore, removal of data outliers is mandatory. However, available statistical techniques are known to lose information at the boundaries of the system and exhibit substantial deviation from holistic data trend. Hence, an adaptive approach combining artificial neural networks and robust winsorization is presented for identification and reconciliation of data outliers in CO2-alkanolamine-H2O system. The proposed approach flexibly transforms to the nonlinear data distribution and predicts corrected values for data outliers (winsorized values), thus maintaining the information at extremes of the system. The results have been graphically analyzed and show good conformance in treated data, with retention of winsorized values. The proposed method improves the shortcomings of previous statistical approaches and can be potentially extended to other nonlinear experimental datasets in chemical process systems.