Episode Mining Device Using Incremental Batch Pattern Segmentation

The patent divides the historical data into multiple time windows (e.g., first time window, second time window, etc.) and processes each window separately to generate candidate patterns. This segmentation allows the system to avoid scanning all historical data repeatedly, instead focusing on specific time segments, thereby reducing processing time while maintaining pattern discovery completeness through systematic coverage of all windows.

The patent performs preliminary processing by pre-dividing historical data into time windows and pre-identifying candidate patterns within each window before final pattern integration. This preliminary action eliminates the need for repeated full-data scanning, as the candidate patterns are already prepared in advance from segmented data, significantly reducing the time required for subsequent pattern matching and validation.

The patent segments the calculation process by computing occurrence frequencies within individual time windows separately rather than scanning all historical data repeatedly. Each time window's candidate patterns are evaluated independently for frequency metrics, and results are aggregated afterward. This segmentation maintains measurement precision through systematic evaluation while reducing computational complexity by avoiding redundant full-data scans.

The patent applies partial action by calculating occurrence frequencies only for candidate patterns identified within specific time windows, rather than evaluating all possible patterns across the entire historical dataset. This approach achieves sufficient measurement precision for the most relevant patterns while significantly reducing computational complexity by focusing resources on partial data subsets that are most likely to contain meaningful patterns.

The patent segments the large dataset into multiple time windows and processes each window independently to generate candidate patterns. This segmentation enables parallel or sequential processing of smaller data subsets, maintaining the quantity of discovered patterns through comprehensive window coverage while significantly improving processing speed by avoiding repeated full-dataset scans. The systematic aggregation of results from all windows ensures pattern quantity is preserved.

The patent performs preliminary pattern identification within each time window before final pattern integration and validation. By pre-processing data in segmented windows and identifying candidate patterns in advance, the system generates comprehensive pattern quantities through systematic coverage while achieving higher processing speeds through optimized, non-repetitive data access and reduced computational overhead.