PARAMETER SET DETERMINATION FOR CLUSTERING OF DATASETS

• US 20170255688A1
• Filed: 03/01/2016
• Published: 09/07/2017
• Est. Priority Date: 03/01/2016
• Status: Active Grant

• Current Assignee: Accenture Global Solutions Limited (Accenture PLC)
• Original Assignee: Accenture Global Solutions Limited (Accenture PLC)
• Inventors: HACKETT-JONES, Emily, CHEW, Theam Yong, KIRKBY, Stephen, PALLOTTA, Rob, ENDERLING, Katherine
• Granted Patent: US 10,235,443 B2
• Time in Patent Office: Days
• CPC Class Codes: G06F 16/285 Clustering or classification
• CPC Class Codes: G06N 20/00 Machine learning
• CPC Class Codes: G06N 3/006 based on simulated virtual individual or collective life forms, e.g. social simulations or particle swarm optimisation [PSO]
• CPC Class Codes: G06N 3/126 Evolutionary algorithms, e.g. genetic algorithms or genetic programming
• CPC Class Codes: G06N 7/01 Probabilistic graphical models, e.g. probabilistic networks

Abstract

A clustering system selects a parameter set from a plurality of parameter sets associated with a dataset to generate a hyper-optimized cluster of the dataset. Different parameter sets are generated by varying the parameter values based on the Genetic Algorithm or a particle swarming algorithm. A parameter set having a high Fitness score is selected from the different parameter sets and a clustered solution produced using the selected parameter set having a maximum total score is used to produce actionable items.

Claims

1. A clustering cycle optimization system comprising:

a data storage to store parameters for a parameter set; and

one or more processors to perform operations to;

generate an optimized parameter set which enables generation of an optimized cluster solution of a dataset by iterating over a plurality of parameter sets;

generate a plurality of cluster solutions from each of a plurality of parameter sets, the plurality of cluster solutions comprising the optimized cluster solution;

select the optimized cluster solution based on a total score of the optimized cluster solution;

select a hyper-optimized parameter set based on its respective fitness score; and

output an actionable item based on the hyper-optimized cluster solution.

2. The clustering cycle optimization system of claim 1, wherein the operations are further to:

generate an initial parameter set of the plurality of parameter sets via random selection of parameter values.

3. The clustering cycle optimization system of claim 2, wherein the operations are further to:

cause a random change in a numerical value of a parameter comprised in the plurality of parameter sets.

4. The clustering cycle optimization system of claim 3, wherein the random change of 5% is caused in the numerical value;

and Boolean and categorical variables are re-sampled.

5. The clustering cycle optimization system of claim 2, wherein the system comprises:

a crossover operator that randomly swaps parameters between two of the plurality of parameter sets.

6. The clustering cycle optimization system of claim 2, wherein the operations are further to:

set categorical distances based on equality of values, wherein non-equal values are equidistant from each other.

7. The clustering cycle optimization system of claim 1, wherein the optimized parameter set has a high fitness score relative to fitness scores of other parameter sets for the same data set.

8. The clustering cycle optimization system of claim 7, wherein the optimized cluster solution has a maximum total score of the plurality of cluster solutions associated with the optimized parameter set.

9. The clustering cycle optimization of claim 7, wherein the fitness score of the current parameter set is an average of total scores of upper quartile subset of the plurality of cluster solutions.

10. The clustering cycle optimization of claim 1, wherein the optimized cluster solution includes clusters of data that each have attributes for determining the actionable item.

11. A method for hyper-optimizing a clustering engine, comprising:

generating a plurality of cluster solutions corresponding to a parameter set of a dataset;

obtaining a Fitness score for the parameter set;

storing the parameter set in a temporary storage;

changing at least one parameter value of the parameter set; and

iterating through the generating, obtaining, storing and changing steps until an average fitness of a current generation of the parameter set has stabilized to within a given percentage of a previous generation'"'"'s average fitness;

selecting a cluster solution having a maximum total score from a plurality of cluster solutions as a hyper-optimized cluster solution, the plurality of cluster solutions generated using the current generation of the parameter set; and

generating an actionable item based on the hyper-optimized cluster solution.

12. The method of claim 11, wherein generating the Fitness score further comprises:

selecting an upper quartile of total scores of the plurality of cluster solutions;

calculating an average of the upper quartile total scores.

13. The method of claim 11, further comprising:

calculating respective total scores for the plurality of cluster solutions wherein each respective total score is a sum of a business value score, a solution diversity score and an average centroid distance score.

14. The method of claim 11, wherein changing at least one parameter value of the parameter set comprises:

perturbing at least one parameter value by a random change.

15. The method of claim 11, wherein changing at least one parameter value of the parameter set comprises:

randomly swapping the parameters of two of the parameter sets.

16. The method of claim 11, wherein changing at least one parameter value of the parameter set comprises:

reselecting categorical parameters randomly from the set of possible values.

17. The method of claim 11, wherein the dataset represents stores and the actionable items comprises:

selecting a subset of the stores for upgrades.

19. The non-transitory computer readable medium of claim 17, wherein the machine readable instructions for changing the parameter value include machine readable instructions to:

perturb the at least one parameter value by a random change.

20. The non-transitory computer readable medium of claim 17, wherein the machine readable instructions for changing the parameter value include machine readable instructions to:

randomly swap the parameters of two of the parameter sets.

18. A non-transitory computer readable medium including machine readable instructions that are executable by at least one processor to:

generate a plurality of clusters corresponding to a parameter set of a dataset;

obtain a Fitness score for the parameter set;

store the parameter set in a temporary storage;

change at least one parameter value of the parameter set;

iterate through the generating, obtaining, storing and changing when a parameter set is generated through the change that has a higher Fitness score until a predetermined number of iterations are performed;

select a cluster having a maximum total score from a plurality of clusters as a hyper-optimized cluster, the plurality of clusters are associated with one parameter set of the predetermined number of parameter sets, the associated parameter set having a high Fitness score; and

perform an actionable item based on the hyper-optimized cluster.