Sales data processing apparatus, method, and medium storing program for sales prediction

A data processing apparatus, comprising

• a processor configured to:
• acquire data including a plurality of records, the data including acceleration data acquired from an acceleration sensor of a mobile sensor;
• first divide the records of the acquired data based on external condition identification information included in each of the records, the external condition identification information identifying which of a plurality of external conditions serves as potential features that affects whether or not a predetermined event occurs, the first division of the records generating data sets for the respective external conditions, and the predetermined event being a user passing through a step as indicated by the acceleration data;
• second divide, for each of the generated data sets for the respective external conditions, records included in the data set based on label information included in each of the records, the label information indicating whether the record corresponds to a positive label indicating that the predetermined event has occurred or a negative label indicating that the predetermined event has not occurred, the second division generating two data sets for the respective label information;
• generate a difference between a first record included in one data set of the two data sets for the respective label information and a second record included in another data set of the two data sets as difference data, wherein the difference data includes a difference of values for respective feature quantities of the respective records and a difference of values indicating whether or not the predetermined event has occurred according to the respective records;
• perform statistical analysis using the generated difference data;
• and
• visually output a result of performing the statistical analysis on a display device;
• and
• a memory for storing instructions for operating the processor,
• wherein each of the records of the acquired data further includes an ID for identifying the user as the external condition identification information, a positive label indicating that the predetermined event has occurred indicates that the user has passed through a step, and a negative label indicating that the predetermined event has not occurred indicates that the user has not passed through the step,
• the difference data further includes a difference of values of the acceleration data according to the respective records, and
• performing the statistical analysis includes calculating a regression coefficient vector according to a model for calculating a score value indicating whether or not the user has passed through a step, by using the difference of the values of the acceleration data included in the difference data as an element of a feature vector and using the difference of the values indicating whether or not the user has passed through the step as a label indicating whether or not the user has passed through the step according to the model, and
• performing learning by adjusting one or more parameters of the statistical analysis based on calculating the regression coefficient vector that maximizes an objective function by approximating a step function using a sigmoid function.

A data processing method executed by an apparatus comprising a hardware processor and a memory, the method comprising:

• acquiring data including a plurality of records, the data including acceleration data acquired from an acceleration sensor of a mobile sensor;
• first dividing the records of the acquired data based on external condition identification information included in each of the records, the external condition identification information identifying which of a plurality of external conditions serves as potential features that affects whether or not a predetermined event occurs, the first dividing the records generating data sets for the respective external conditions, and the predetermined event being a user passing through a step as indicated by the acceleration data;
• second dividing, for each of the generated data sets for the respective external conditions, records included in the data set based on label information included in each of the records, the label information indicating whether the record corresponds to a positive label indicating that the predetermined event has occurred or a negative label indicating that the predetermined event has not occurred, the second dividing generating two data sets for the respective label information, the positive label indicating that the predetermined event has occurred and that the user has passed through a step, and the negative label indicating that the user has not passed through a step;
• generating a difference between a first record included in one data set of the two data sets for the respective label information and a second record included in another data set of the two data sets as difference data, wherein the difference data includes a difference of values for respective feature quantities of the respective records and a difference of values indicating whether or not the predetermined event has occurred according to the respective records;
• performing statistical analysis using the generated difference data;
• and
• visually outputting a result of performing the statistical analysis on a display device,
• wherein each of the acquired records further includes information corresponding to a value of a predetermined feature quantity,
• performing the statistical analysis includes calculating a regression coefficient vector according to a model for calculating a score value indicating a likelihood of an occurrence of the predetermined event by using the difference of the values for the respective feature quantities of the predetermined feature quantity included in the difference data as an element of a feature vector and using the difference of the values indicating whether or not the predetermined event has occurred included in the difference data as a label indicating whether or not the predetermined event has occurred according to the model,
• and performing learning by adjusting one or more parameters of the statistical analysis based on calculating the regression coefficient vector that maximizes an objective function by approximating a step function using a sigmoid function.

A data processing method executed by an apparatus comprising a hardware processor and a memory, the method comprising:

• acquiring data including a plurality of records, the data including acceleration data acquired from an acceleration sensor of a mobile sensor;
• first dividing the records of the acquired data based on external condition identification information included in each of the records, the external condition identification information identifying which of a plurality of external conditions serves as potential features that affects whether or not a predetermined event occurs, the first dividing the records generating data sets for the respective external conditions, and the predetermined event being a user passing through a step as indicated by the acceleration data;
• second dividing, for each of the generated data sets for the respective external conditions, records included in the data set based on label information included in each of the records, the label information indicating whether the record corresponds to a positive label indicating that the predetermined event has occurred or a negative label indicating that the predetermined event has not occurred, the second dividing generating two data sets for the respective label information;
• generating a difference between a first record included in one data set of the two data sets for the respective label information and a second record included in another data set of the two data sets as difference data, wherein the difference data includes a difference of values for respective feature quantities of the respective records and a difference of values indicating whether or not the predetermined event has occurred according to the respective records;
• performing statistical analysis using the generated difference data;
• and
• visually outputting a result of performing the statistical analysis on a display device,
• wherein each of the records of the acquired data further includes an ID for identifying the user as the external condition identification information, a positive label indicating that the predetermined event has occurred indicates that the user has passed through a step, and a negative label indicating that the predetermined event has not occurred indicates that the user has not passed through the step,
• the difference data further includes a difference of values of the acceleration data according to the respective records,
• performing the statistical analysis includes calculating a regression coefficient vector according to a model for calculating a score value indicating whether or not the user has passed through a step, by using the difference of the values of the acceleration data included in the difference data as an element of a feature vector and using the difference of the values indicating whether or not the user has passed through the step as a label indicating whether or not the user has passed through the step according to the model, and
• performing learning by adjusting one or more parameters of the statistical analysis based on calculating the regression coefficient vector that maximizes an objective function by approximating a step function using a sigmoid function.

A non-transitory tangible computer-readable storage medium storing instructions for causing a processor to:

• acquire data including a plurality of records, the data including acceleration data acquired from an acceleration sensor of a mobile sensor;
• first divide the records of the acquired data based on external condition identification information included in each of the records, the external condition identification information identifying which of a plurality of external conditions serves as potential features that affects whether or not a predetermined event occurs, the first division of the records generating data sets for the respective external conditions, and the predetermined event being a user passing through a step as indicated by the acceleration data;
• second divide, for each of the generated data sets for the respective external conditions, records included in the data set based on label information included in each of the records, the label information indicating whether the record corresponds to a positive label indicating that the predetermined event has occurred or a negative label indicating that the predetermined event has not occurred, the second division generating two data sets for the respective label information, the positive label indicating that the user has passed through a step and the negative label indicating that the user has not passed through a step;
• generate a difference between a first record included in one data set of the two data sets for the respective label information and a second record included in another data set of the two data sets as difference data, wherein the difference data includes a difference of values for respective feature quantities of the respective records and a difference of values indicating whether or not the predetermined event has occurred according to the respective records;
• perform statistical analysis using the generated difference data;
• and
• visually output a result of performing the statistical analysis on a display device,
• wherein each of the records of the acquired data further includes information corresponding to a value of a predetermined feature quantity,
• performing the statistical analysis includes calculating a regression coefficient vector according to a model for calculating a score value indicating a likelihood of an occurrence of the predetermined event by using the difference of the values for the respective feature quantities of the predetermined feature quantity included in the difference data as an element of a feature vector and using the difference of the values indicating whether or not the predetermined event has occurred included in the difference data as a label indicating whether or not the predetermined event has occurred according to the model, and
• perform learning by adjusting one or more parameters of the statistical analysis based on calculating the regression coefficient vector that maximizes an objective function by approximating a step function using a sigmoid function.

A non-transitory tangible computer-readable storage medium storing instructions for causing a processor to:

• acquire data including a plurality of records, the data including acceleration data acquired from an acceleration sensor of a mobile sensor;
• first divide the records of the acquired data based on external condition identification information included in each of the records, the external condition identification information identifying which of a plurality of external conditions serves as potential features that affects whether or not a predetermined event, the first division of the records generating data sets for the respective external conditions, and the predetermined event being a user passing through a step as indicated by the acceleration data;
• second divide, for each of the generated data sets for the respective external conditions, records included in the data set based on label information included in each of the records, the label information indicating whether the record corresponds to a positive label indicating that the predetermined event has occurred or a negative label indicating that the predetermined event has not occurred, the second division generating two data sets for the respective label information;
• generate a difference between a first record included in one data set of the two data sets for the respective label information and a second record included in another data set of the two data sets as difference data, wherein the difference data includes a difference of values for respective feature quantities of the respective records and a difference of values indicating whether or not the predetermined event has occurred according to the respective records;
• perform statistical analysis using the generated difference data;
• and
• visually output a result of performing the statistical analysis on a display device,
• wherein each of the records of the acquired data further includes an ID for identifying the user as the external condition identification information, a positive label indicating that the predetermined event has occurred indicates that the user has passed through a step, and a negative label indicating that the predetermined event has not occurred indicates that the user has not passed through the step,
• the difference data further includes a difference of values of the acceleration data according to the respective records, and
• perform the statistical analysis includes calculating a regression coefficient vector according to a model for calculating a score value indicating whether or not the user has passed through a step, by using the difference of the values of the acceleration data included in the difference data as an element of a feature vector and using the difference of the values indicating whether or not the user has passed through the step as a label indicating whether or not the user has passed through the step according to the model, and
• perform learning by adjusting one or more parameters of the statistical analysis based on calculating the regression coefficient vector that maximizes an objective function by approximating a step function using a sigmoid function.