Systems and models for data analytics
View Patent ↗Systems and methods are provided that allow for generating and applying an improved predictive data model that aggregates two or more models performed sequentially, for the purposes of improving the prediction of overall profitability of individuals or households in a population. The models may be generated by the processing of customer profitability data and third-party population data together. One of the two aggregated models may be an inherently probabilistic, binary model tasked with determining whether an individual is a high-loss individual and using that result to improve the predictive capability of the system.
1. A computing system comprising:
one or more data stores storing:
a first dataset including first data items associated with respective individuals of a first plurality of individuals; and
a second dataset including second data items associated with respective individuals of at least some of the first plurality of individuals, wherein the second data items include known profitability values associated with respective individuals;
a computer processor; and
a computer readable storage medium storing program instructions configured for execution by the computer processor in order to cause the computer processor to:
access the first dataset and the second dataset;
generate a training dataset by at least:
performing a fuzzy match between the first dataset and the second dataset to identify a plurality of overlapping individuals associated with both the first dataset and the second dataset;
for each individual of the plurality of overlapping individuals:
adding, to the training dataset, first data items associated with the individual; and
adding, to the training dataset, second data items associated with the individual;
randomly divide the training dataset into a first subset of data items associated with respective of a first subset of the plurality of overlapping individuals, and a second subset of data items associated with respective of a second subset of the plurality of overlapping individuals;
generate a first predictive model configured to determine a predicted profitability of an individual by at least:
training the first predictive model based on the first subset of data items associated with respective of the first subset of the plurality of overlapping individuals; and
validating the first predictive model by comparing predicted profitabilities of individuals of the second subset of the plurality of overlapping individuals with known profitability values of individuals of the second subset of the plurality of overlapping individuals;
generate a second predictive model configured to determine a predicted likelihood of disaster of an individual by at least:
training the second predictive model based on the first subset of data items associated with respective of the first subset of the plurality of overlapping individuals; and
validating the second predictive model by comparing predicted likelihoods of disaster of individuals of the second subset of the plurality of overlapping individuals with known profitability values of individuals of the second subset of the plurality of overlapping individuals;
access a third dataset including third data items associated with a third plurality of individuals;
apply the first predictive model to the third dataset to determine predicted profitabilities of respective individuals of the third plurality of individuals;
apply the second predictive model to the third dataset to determine predicted likelihoods of disaster of respective individuals of the third plurality of individuals;
filter, based on the predicted likelihoods of disaster of respective individuals of the third plurality of individuals, the third dataset to determine a subset of the third plurality of individuals that are unlikely to experience a disaster;
determine an order of the subset of the third plurality of individuals based on their respective predicted profitabilities; and
present a list of the subset of the third plurality of individuals that is ordered according to the order.
2. The computing system of claim 1 , wherein the first data items in the first dataset comprise demographic data, geographical data, and/or behavioral data associated with respective individuals of the first plurality of individuals.
3. The computing system of claim 1 , wherein training the first predictive model further comprises using a decision tree algorithm to perform a segmented linear regression on the first subset of data items associated with respective of the first subset of the plurality of overlapping individuals, and wherein training the second predictive model further comprises using a decision tree algorithm to perform a segmented linear regression on the first subset of data items associated with respective of the first subset of the plurality of overlapping individuals.
4. The computing system of claim 1 , wherein the third data items in the third dataset comprise demographic data, geographical data, or behavioral data associated with the third plurality of individuals.
5. The computing system of claim 1 , wherein the plurality of overlapping individuals of the training dataset is fewer individuals than the first plurality of individuals in the first dataset.
6. The computing system of claim 1 , wherein performing a fuzzy-match between the first dataset and the second dataset to identify a plurality of overlapping individuals further comprises comparing an identifier of each individual in the first dataset to an identifier of each individual in the second dataset.
7. The computing system of claim 6 , wherein the identifier comprises at least one of: a name associated with an individual, an address associated with an individual, a postal code associated with an individual, and a phone number associated with an individual.
8. The computing system of claim 1 , wherein the first subset of data items of the training dataset is greater than approximately half of the data items in the training dataset, and wherein the second subset of data items of the training dataset is less than approximately half of the data items in the training dataset.
9. The computing system of claim 1 , wherein the second predictive model is a high-loss binary model.
10. The computing system of claim 1 , wherein the second predictive model is further configured to calculate the average likelihood of disaster in a group of individuals and compare it to the predicted likelihood of disaster of an individual in order to classify the individual as either a high-loss individual or a small-loss individual.
11. The computing system of claim 1 , wherein the second predictive model is further configured to use the predicted likelihood of disaster of an individual in order to classify each individual as either a high-loss individual or a small-loss individual.
12. The computing system of claim 11 , wherein individuals classified as small-loss individuals are unlikely to experience a disaster, and wherein individuals classified as high-loss individuals are likely to experience a disaster.
13. The computing system of claim 12 , wherein filtering the third dataset to determine a subset of the third plurality of individuals that are unlikely to experience a disaster further comprises: determining a subset of the third plurality of individuals that are classified as high-loss individuals and likely to experience a disaster and removing that subset of high-loss individuals from the third plurality of individuals.
14. The computing system of claim 13 , wherein the program instructions further cause the computing system to:
generate data useable for rendering a user interface including a plot including:
a Y-axis representing an average of the predicted profitabilities of the subset of the third plurality of individuals that are small-loss individuals and unlikely to experience a disaster, and
an X-axis representing a percentage of individuals of the third plurality of individuals that have been classified as high-loss individuals and filtered from the third plurality of individuals.
15. The computing system of claim 1 , wherein the first predictive model is further configured to determine a predicted profitability of an individual using demographic data, geographical data, and/or behavioral data associated with the individual.
16. The computing system of claim 1 , wherein performing a fuzzy match between the first dataset and the second dataset further comprises normalizing values of the first data items of the first dataset to be within zero and one.
17. The computing system of claim 1 , wherein the program instructions further cause the computing system to sort the subset of the third plurality of individuals into a plurality of profitability ranges based the predicted profitabilities of the respective individuals.
18. The computing system of claim 17 , wherein the program instructions further cause the computing system to:
generate data useable for rendering a user interface including a map that visually presents, on the map:
a map of a geographical region associated with the third plurality of individuals; and
a plurality of tiles overlaid on the map, wherein each tile is associated with a respective individual of the third plurality of individuals, and wherein each tile is colored based on the profitability range that the predicted profitability of the respective individual is in.
19. The computing system of claim 1 , wherein the program instructions further cause the computing system to:
generate data useable for rendering a user interface including a map that visually presents, on the map, a predicted profitability of at least one individual of the subset of the third plurality of individuals based on a corresponding address or location of the at least one individual.
20. The computing system of claim 1 , wherein the first subset of data items in the training dataset and the second subset of data items in the training dataset are non-overlapping.