Increasing index sensitivity Clause Samples

Increasing index sensitivity. We next considered expansion of the “best usage” statistic of a cluster to other usages that performed well. Table 4.4 shows the percent of data sets for which the highest-ARI partition was chosen by any statistic of an index that chose correctly more than 20% of the time. In several cases, inclusion of other usages increased the probability of identifying the highest-ARI partition. We chose the ▇▇▇▇▇▇-▇▇▇▇▇▇▇, ▇▇▇▇▇▇-▇▇▇▇▇▇▇ (SAD), Xu, and Xu (SAD) indices for further evaluation. These indices had several usages that performed well overall. Combining the best usages improved the sensitivity of the index/usage combinations, or the probability of one of the best usages of the index to choose the best partition. Either the minimum difference to the left or the maximum second difference (or both) of the Xu and Xu (SAD) indices chose the best partition in 70.5% and 69.8% of cases. For the ▇▇▇▇▇▇- ▇▇▇▇▇▇▇ and ▇▇▇▇▇▇-▇▇▇▇▇▇▇ (SAD) indices, 79.7% and 78.3% of the best partitions were chosen by either the maximum second difference, the the maximum difference to the right, or the global minimum.