Clearly, models can be no better than the data on which they are based. Considering the limitations of the data is an important step in constructing robust and reliable Quantitative Structure-Property Relationships (QSPR). Using ever better interpolation techniques may lead to models that are superficially more accurate than the data on which they are based. This is clearly undesirable but depressingly common.<br/><br/>To turn the question around; what do the models tell us about the data? Can we gain additional information about the quality and applicability domain of the training data by analyzing the performance of the model in detail? Real examples will be given in addition to simulated models using artificial noise in the training data.

Data and models, models and data

249th National Meeting (2015)

The official archive and repository for all ACS Meetings & Expositions abstracts and select posters or presentations.