Figure 8.
The K-means algorithm. The K-means algorithm was applied for validating the application of a representative set synergies, using the full activation coefficients (H matrices). For each group the cross validation procedure yielded 81 full coefficient matrices (H). A constant set of features (see also method section) was extracted from each matrix. Each of the H matrices features was calculated as a data point (a single row), which composed of 44 features (11 features × 4 synergies). The K-means algorithm was iterated 10 times, changing the number of clusters/ centroids (K) from four to nine. The clustering indices were ordered to be aligned with the cross validation matrix (Figure 3) as illustrated in (A). Each row was assigned to a cluster index value given by the MATLAB algorithm, according to the most common index in a row. The assigned indices are shown in the legend bar (A). The accuracy of classification was computed using the purity index (see also method section) for each running of the algorithm with each K's. In (A) for example the purity was calculated as follow: ((8 + 9 + 9 + 9 + 6 + 8 + 9 + 7 + 9)/81)*100, which resulted in 91.358% accuracy. (B) Illustrates the location of the indices ordered according to the target directions as in Figure 1B. (C) The average accuracy from the 10 iterations for each K was computed and plotted, for each group separately.