Hyperlift Analysis
The Hyperlift approach is a visual method to evaluate the correlation between two variables using CrossRank technology (cf the article What is the Crossrank?). This presentation delivers all dataset characteristics at a glance by comparing an input variable's distribution against a target output.
Visual Components:
- Green Bars: Indicate a positive impact on your output within a specific range.
- Red Bars: Indicate a negative impact on your output within a specific range.
- Green Area (Background): Represents the "Best Range," identifying the optimal operating zone for your output.
- Blue Histogram (Bottom): Displays the distribution (frequency) of the input variable.
The Math: "Problem Solving" through Lift
The Lift value represents the percentage of a "problem" solved by reducing the ratio of "Bad" (Red) points in a specific subpopulation. The shift is always calculated based on the population of the opposite color.
- Logic: A Green Lift indicates a shift from Red (Bad) to Green (Good).
- Example (Model with 60% Red / 40% Green baseline):
- In a subpopulation with a 50% Green Lift, the calculation is: 50% x 60%(initial Red points)} = 30% of the total subpopulation shifts to Green.
- Final Result: 30% Red (60% - 30%) and 70% Green (40% + 30%).
- In this case, 30% of the overall problem is considered "solved" within this range.
Important Algorithmic Notes:
- Variable Filtering: The algorithm automatically removes variables with a very low CrossRank.
- Zero Visibility: If all activated variables have a CrossRank of 0, the software will not display any hypercubic rules.