3.2 Experimental Results
In the simulation data set experiment, the detection model based on Kolmogorov–
Smirnov test is used to detect whether the two data blocks obey the same potential
distribution. The horizontal red line indicates that the significant level of a is 0.05. The
results show that whether the difference between the data blocks is less than 0.05 or
greater than 0.05. If the p-value between the data blocks is greater than 0.05, we will
accept the initial assumption that the data blocks before and after conform to the same
potential distribution and it means no concept drift occurs between the two data blocks.
But if the p-value is less than 0.05, we will turn down the initial assumption as the data
blocks before and after do not conform to the same potential distribution, which states
clearly that there is an obvious difference between the data blocks before and after, that
is, concept drift has occurred. By comparing the significance level a and p-value of
simulation data, we can clearly see that the model can accurately detect whether there is
a significant difference between the data blocks before and after (Fig. 1).
In terms of radar data that contains some noise, by comparing Fig. 2 and result
plots, we can see that errors occurred at 5 and 7 when testing the significance level, but
as for its overall results, the K–S test can accurately detect whether there is concept
drift between adjacent data blocks. And when a is 0.05, the result is highly reliable,
therefore K–S detection can be used to detect whether concept drift occurs in the data
stream.
Fig. 1. Simulation data result
278
Z. Wang and W. Wang
In the simulation data set experiment, the detection model based on Kolmogorov–
Smirnov test is used to detect whether the two data blocks obey the same potential
distribution. The horizontal red line indicates that the significant level of a is 0.05. The
results show that whether the difference between the data blocks is less than 0.05 or
greater than 0.05. If the p-value between the data blocks is greater than 0.05, we will
accept the initial assumption that the data blocks before and after conform to the same
potential distribution and it means no concept drift occurs between the two data blocks.
But if the p-value is less than 0.05, we will turn down the initial assumption as the data
blocks before and after do not conform to the same potential distribution, which states
clearly that there is an obvious difference between the data blocks before and after, that
is, concept drift has occurred. By comparing the significance level a and p-value of
simulation data, we can clearly see that the model can accurately detect whether there is
a significant difference between the data blocks before and after (Fig. 1).
In terms of radar data that contains some noise, by comparing Fig. 2 and result
plots, we can see that errors occurred at 5 and 7 when testing the significance level, but
as for its overall results, the K–S test can accurately detect whether there is concept
drift between adjacent data blocks. And when a is 0.05, the result is highly reliable,
therefore K–S detection can be used to detect whether concept drift occurs in the data
stream.
Fig. 1. Simulation data result
278
Z. Wang and W. Wang