Mimir Analytics
Classification Modeling
​
When you need a classification algorithm, reach out to us. We generate models to classify outcomes, such as purchase vs. no-purchase, within days of receiving your data. We currently offer logistic regression and classification tree algorithms. All our algorithms come with box plots and distribution metrics that underlie the features that enter the models.
​
The image below gives an example of how one of our logistic regression models captures the percent of events over the test data by vigintile. This model captured 83.75% of all the events in 5% of the scored file.
Our algorithms are easy to implement. The algorithm that generated the above segments is:
Logistic_Score = 8.093 - 9.586*V3_scld*V14_scld - 11.565*V12_scld*V14_scld
​
-
In the above scoring function we have scaled the raw data and identified significant interactions.
-
We also provide our clients with precision vs. recall (TPR) thresholds.
The image below is an example of a classification tree. There is no need to manually code the tree logic to implement this model. We provide tree path text that can easily by transformed into script and hence production.
This tree has the following characteristics:
​
Accuracy (test): 0.90
Precision (test): 0.84
Recall (test): 0.91
F1 score (test): 0.88
Features in Tree:
['V3', 'V1', 'V2']
Tree Paths:
Path_1: WHEN V1 <= 0.3202 AND V3 > -3.9649 THEN 0
Path_2: WHEN V1 <= 0.3202 AND V3 <= -3.9649 THEN 1
Path_3: WHEN V1 <= 0.3202 AND V2 <= 7.5653 THEN 1
Path_4: WHEN V1 <= 0.3202 AND V2 > 7.5653 THEN 0
​
Contact us to receive a sample report of a model build and the underlying data structure.
CONTACT US:
Michael E Seidel, M.Sc.(Econ), CFA
Proprietor & Chief Data Scientist
mseidel@mimiranalytics.com
818-517-6928
Econometrician with twenty five years of experience in various industries. Including:
-
Club membership programs
-
Healthcare
-
Veterinary Care
-
Financial Services (Mortgages and Insurance)
-
Banking — Payment Processing
-
Home Video
​