Device and method for multi-class classification by machine learning

1. A device for multi-class classification of an application relative to a classification of a transport mode by machine learning, the device comprising:

sensors configured to acquire, at successive instants, measurements of physical quantities specific to said application; and

a microprocessor configured to:

acquire said successive measurements from the sensors to calculate predictors from physical quantities of said measurements, said predictors being indicative of a position or of a state of mobility,

construct continuous series of samples, each sample being a vector formed of a predetermined number of predictors, break down each series of samples into a set of distinctive sequences each corresponding to a specific unique class among a set of predetermined classes, said set of predetermined classes corresponding to different transport modes,

under-sample the sequences associated with the unique classes, without under-sampling the sequences associated with an overall class, among said set of predetermined classes while conserving for each selected sequence a predetermined fraction of samples of said sequence, thereby forming classes associated with balanced samples, and

construct a classifier defining the transport mode classification from said balanced samples by machine learning.

2. The device according to claim 1 , wherein the predetermined fraction of samples comprises samples at limits of the each selected sequence and intermediate samples.

3. The device according to claim 1 , wherein intermediate samples are regularly spaced apart in said each selected sequence by selecting a sample out of a predetermined number of samples.

4. The device according to claim 1 , wherein the under-sampled sequences are sequences associated with a sub-set of majority classes among said set of predetermined classes.

5. The device according to claim 4 , wherein a majority class is a class of which the number of samples is greater than an average value of samples of the different classes.

6. The device according to claim 4 , wherein a majority class is a class of which cumulative number of samples represents a predetermined percentage of a total number of samples of the different classes.

7. The device according to claim 1 , wherein the microprocessor is further configured to select the sequences to under-sample among said set of distinctive sequences as a function of predetermined selection criteria.

8. The device according to claim 7 , wherein the predetermined selection criteria comprise the selection of sequences having a similarity measurement greater than a predetermined threshold.

9. The device according to claim 8 , wherein the similarity measurement of a sequence corresponds to the average of the similarity measurements determined on each of the predictors of the sequence.

10. The device according to claim 7 , wherein the predetermined selection criteria comprise the selection of sequences having a duration greater than a predetermined temporal period.

11. The device according to claim 1 , wherein the microprocessor is configured to determine a carbon footprint of a user of said device as a function of said classification of the transport mode.

12. The device according to claim 1 , wherein

the device is a predictive analysis device, and

the predictive analysis device is a mobile telephone.

13. A non-transitory computer readable medium storing instructions which when executed by a processor cause the processor to perform a method for multi-class classification of an application relative to a classification of a transport mode by machine learning, the method comprising:

acquiring successive temporal measurements of physical quantities specific to said application,

calculating predictors from said physical quantities, said predictors being indicative of a position or of a state of mobility,

constructing from successive measurements series of samples, each sample being a temporal vector formed of a predetermined number of predictors,

breaking down each series of samples into a set of distinctive sequences each corresponding to a specific unique class among a set of predetermined classes, said set of predetermined classes corresponding to different transport modes,

under-sampling sequences associated with the unique classes, without under-sampling the sequences associated with an overall class, among said set of predetermined classes while conserving for each selected sequence a predetermined fraction of samples of said sequence, thereby forming classes associated with balanced samples, and

constructing a classifier defining the transport mode classification from said balanced samples by machine learning.