Machine learning can be used as an effective analysis tool complementary to physical modelling also on relatively small-size datasets. The fact that data was not specifically collected with this idea in mind can have significant implications on which methods are appropriate, but important insights can nevertheless be extracted with surprisingly little effort - which may translate to lower costs. In my presentation, I substantiate these claims by analyzing data which was provided by Roche Diagnostics International and was acquired during an end-of-life study on electrochemical sensors for a blood gas electrolytes instrument.