When overfilting the data, researchers are making a classification based on such a small amount of data, that they may be:

fitting to the noise $($the randomness$)$ in the data while ignoring intuitive information about the more typical outcomes of historical recor ous areas of a scatter plot.

magnifying the intuitive information about typical outcomes of historical records in the various areas of a scatter plot.

reducing the ability to smooth the data and develop a line of best fit.