Best k for k-NN

We tried this experiment for SIFT with topology: 15x9x25. This one was the best so far. Here we can see how the k parameter influences the classification itself. We can see that with increasing k the classification rate is decreasing (figure 4.14 on page ). The best classification rate was obtained for $k=1$ however the classification rate remains almost the same for all tried $k$.

Table 4.9: Best k for k-NN and how this influence the classification rate(see figure 4.14 on page )

k for k-NN	Classification Rate [%]
1	92.39
3	86.95
5	90.21
7	85.86
9	88.04
11	85.86
13	84.78
15	84.78
17	84.78
19	83.69
21	85.86
23	83.69

Figure 4.14: The influence of k on classification rate