Multiple sets of data sets, the accuracy of the last set cannot be improved

[babyta]

hello author，I used 9 sets of data, using the incremental learning method, starting from one set and slowly increasing it. Later, when I added DIS data, the reorganized data verification accuracy was lower, as shown below
_maxF1_0.9685_0.9713_0.9704_0.9713_0.9703_0.9729_0.9709_0.8183_0.9881
0.8183 is the accuracy of the DIS data set. When I use this data alone for training, the accuracy is very high. I suspect it is a problem with the distribution of the rest of my self-built data and the current data.