Figure 10. Classification success of models in different datasets
In the final stage of the performance evaluation of the proposed model,
the data set from three locations and collected in the central server
and the weights of the trained model were transferred to the edge
systems and evaluated. Here, it has been observed that the models in the
locations have never been encountered before, and that the data ranges
can be classified by combining the weights obtained from the training of
other locations. First of all, the weights of the main model trained
with the central dataset were distributed to the edge locations and the
weight was updated. Afterwards, each location was asked to classify the
data ranges they encountered for the first time, as shown in Table 5. As
shown in Figure 11, data ranges encountered for the first time were
recognized by the weights learned in the main model, and order
classification was performed with an average of 95% accuracy.