Portfolio Matching Control Location Selection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for determining control locations in test versus control analysis, such as the group-to-group and similar sites approaches, suffer from measurement errors due to suboptimal baseline selection and computational inefficiencies in identifying the most suitable control sites for each test site.
Innovation Solution
A system and method that uses a hill-climbing algorithm to optimize the selection of a portfolio of control locations by defining an objective function to score the similarity of control locations averaged together, efficiently identifying an optimal set of control locations that collectively match the historical performance of a test site, thereby reducing measurement errors.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If the similar sites approach is used to select N control locations individually most similar to a test location, then measurement precision is improved, but the selected control locations may be collectively biased in the same direction, reducing reliability
Solution Approach 1:
The patent changes the optimization parameter from individual site similarity to portfolio-level similarity. Instead of selecting N control sites that individually maximize similarity metrics, the system selects sites that collectively minimize the difference between the test site's historical performance and the average performance of the control portfolio, ensuring both precision and reliability
Solution Approach 2:
The patent inverts the traditional selection approach by not directly selecting sites based on their individual similarity scores, but rather by selecting sites whose average performance best matches the test site. This inversion transforms the selection criterion from individual optimization to collective optimization, resolving the bias problem
2Measurement precision
If all possible combinations of control locations are evaluated to find the optimal portfolio, then measurement precision is improved, but computational resources required become prohibitively large
Solution Approach 1:
The patent segments the combinatorial optimization problem into manageable components by using an iterative swapping approach. Instead of evaluating all combinations, it starts with an initial portfolio and makes incremental improvements by swapping individual sites, reducing computational complexity from factorial to polynomial scale while maintaining precision
Solution Approach 2:
The patent applies a dynamic optimization process where the control portfolio is iteratively improved through successive swaps. The system dynamically adjusts the portfolio composition by evaluating marginal improvements from swapping individual sites, allowing efficient convergence to an optimal solution without exhaustive search
Data Source
AI summary
The embodiments the systems and methods described herein attempt to optimally select a group or portfolio of control locations for each test location. The optimization can be generally performed in two steps. First, an objective function is defined that scores the similarity of a set of control locations averaged together. Second, given the large number of potential solutions, a computationally-feasible algorithm that identifies an optimal set of control locations and is based on the objective function is executed. In order to obtain the optimal set of control locations in an efficient manner for use in business analytics, the algorithm may use a hill-climbing algorithm. As a result, an optimization function can be incrementally improved in an efficient manner.