Pattern Determination Device for High-Dimensional Clustering

The patent changes the parameter of distance metric from conventional L2 norm to L1/k norm (where k≥2). This parameter change fundamentally alters how distances are calculated in high-dimensional space, preventing spherical surface concentration and maintaining stable neighbor determination. The L1/k norm transforms the distance calculation to be more robust in high dimensions, directly resolving the contradiction between maintaining accuracy and adapting to high-dimensional spaces.

The patent converts the harmful effect of outliers and missing values into a beneficial feature by using the L1/k norm. This metric inherently reduces the influence of large deviations (outliers) and handles missing values gracefully, eliminating the need for complex pre-processing. The harm of noisy data is transformed into a robustness advantage, resolving the contradiction between improving reliability and reducing complexity.

The patent introduces a new parameter k (where k≥2) to modify the L1 norm, creating the L1/k norm. This parameter adjustment fine-tunes the balance between robustness and accuracy: the norm power 1/k reduces outlier influence more gently than standard L1, maintaining better neighbor determination accuracy in high-dimensional spaces while preserving robustness against outliers.

Instead of reducing dimensions, the patent changes the distance metric parameter to L1/k norm, allowing direct operation in the original high-dimensional space. This parameter change preserves all original features while making the distance calculation suitable for high dimensions, avoiding information loss from dimension reduction while maintaining computational feasibility.