Causal Inference Machine Learning with Statistical Background Subtraction

Abstract

A system and method are disclosed to generate causal inference machine learning models employing statistical background subtraction. Embodiments include a server comprising a processor and memory. Embodiments receive historical sales data for one or more past time periods and corresponding historical data for one or more causal variables. Embodiments deconfound the cause-effect relationship of historical sales data and historical data on the one or more causal variables. Embodiments define one or more sample weights for statistical background subtraction of the historical data and perform statistical background subtraction on the historical data. Embodiments train a first machine learning model to predict an absolute individual causal effect on a considered demand quantity in relation to the one or more causal variables and one or more sample weights.

What is claimed is: 
     
         1 . A computer-implemented method, comprising:
 receiving, with a server comprising a processor and memory, historical sales data for one or more past time periods and corresponding historical data for one or more causal variables;   deconfounding, by the server, a cause-effect relationship of historical sales data and historical data on the one or more causal variables by conducting one or more randomized controlled A/B group trials;   defining, by the server, one or more sample weights for statistical background subtraction of the historical data;   performing, by the server, statistical background subtraction on the historical data;   training, by the server, by an iterative approach comprising cyclic boosting in additive regression mode, a first machine learning model to predict an absolute individual causal effect on a considered demand quantity; and   predicting, by the server, with the first machine learning model, an absolute individual causal effect on one or more considered demand quantities during a prediction period by training a second machine learning model.   
     
     
         2 . The computer-implemented method of  claim 1 , wherein a casual variable comprises a sending of a personalized coupon. 
     
     
         3 . The computer-implemented method of  claim 1 , further comprising:
 restricting, by the server, a learning of a causal factor between a feature describing seasonality over a year and a target to a smooth sinusoidal dependency.   
     
     
         4 . The computer-implemented method of  claim 1 , further comprising:
 assigning, by the server, one or more positive sample weights and one or more negative sample weights to one or more samples of the one or more randomized controlled A/B group trials.   
     
     
         5 . The computer-implemented method of  claim 1 , further comprising:
 predicting, by the server, one or more what-if volume predictions from one or more sets of hypothetical causal factors.   
     
     
         6 . The computer-implemented method of  claim 1 , further comprising:
 predicting, by the server, one or more individual causal effects on gross margin.   
     
     
         7 . The computer-implemented method of  claim 1 , further comprising:
 modelling, by the server, demand as a negative binomial or Poisson-Gamma distribution.   
     
     
         8 . A system, comprising:
 a server comprising a processor and memory and configured to:
 receive historical sales data for one or more past time periods and corresponding historical data for one or more causal variables; 
 deconfound a cause-effect relationship of historical sales data and historical data on the one or more causal variables by conducting one or more randomized controlled A/B group trials; 
 define one or more sample weights for statistical background subtraction of the historical data; 
 perform statistical background subtraction on the historical data; 
 train, by an iterative approach comprising cyclic boosting in additive regression mode, a first machine learning model to predict an absolute individual causal effect on a considered demand quantity; and 
 predict, with the first machine learning model, an absolute individual causal effect on one or more considered demand quantities during a prediction period by training a second machine learning model. 
   
     
     
         9 . The system of  claim 8 , wherein a casual variable comprises a sending of a personalized coupon. 
     
     
         10 . The system of  claim 8 , wherein the server is further configured to:
 restrict a learning of a causal factor between a feature describing seasonality over a year and a target to a smooth sinusoidal dependency.   
     
     
         11 . The system of  claim 8 , wherein the server is further configured to:
 assign one or more positive sample weights and one or more negative sample weights to one or more samples of the one or more randomized controlled A/B group trials.   
     
     
         12 . The system of  claim 8 , wherein the server is further configured to:
 predict one or more what-if volume predictions from one or more sets of hypothetical causal factors.   
     
     
         13 . The system of  claim 8 , wherein the server is further configured to:
 predict one or more individual causal effects on gross margin.   
     
     
         14 . The system of  claim 8 , wherein the server is further configured to:
 model demand as a negative binomial or Poisson-Gamma distribution.   
     
     
         15 . A non-transitory computer-readable storage medium embodied with software, the software when executed configured to:
 receive, with a server comprising a processor and memory, historical sales data for one or more past time periods and corresponding historical data for one or more causal variables;   deconfound a cause-effect relationship of historical sales data and historical data on the one or more causal variables by conducting one or more randomized controlled A/B group trials;   define one or more sample weights for statistical background subtraction of the historical data;   perform statistical background subtraction on the historical data;   train, by an iterative approach comprising cyclic boosting in additive regression mode, a first machine learning model to predict an absolute individual causal effect on a considered demand quantity; and   predict, with the first machine learning model, an absolute individual causal effect on one or more considered demand quantities during a prediction period by training a second machine learning model.   
     
     
         16 . The non-transitory computer-readable storage medium of  claim 15 , wherein a casual variable comprises a sending of a personalized coupon. 
     
     
         17 . The non-transitory computer-readable storage medium of  claim 15 , wherein the software when executed is further configured to:
 restrict a learning of a causal factor between a feature describing seasonality over a year and a target to a smooth sinusoidal dependency.   
     
     
         18 . The non-transitory computer-readable storage medium of  claim 15 , wherein the software when executed is further configured to:
 assign one or more positive sample weights and one or more negative sample weights to one or more samples of the one or more randomized controlled A/B group trials.   
     
     
         19 . The non-transitory computer-readable storage medium of  claim 15 , wherein the software when executed is further configured to:
 predict one or more what-if volume predictions from one or more sets of hypothetical causal factors.   
     
     
         20 . The non-transitory computer-readable storage medium of  claim 15 , wherein the software when executed is further configured to:
 predict one or more individual causal effects on gross margin.