Machine learning-based, predictive, digital underwriting system, digital predictive process and corresponding method thereof

1. A machine-learning (ML) based, predictive, digital underwriting system providing an automated parameter-driven predictive underwriting process based on measured probability values associated with individuals of a cohort or portfolio, the individuals being exposed to a probability of occurrence of one or more predefined medical and/or health and/or life events having the probability value with a predefined severity within a future measuring time-window, the ML-based, predictive, digital underwriting system comprising:

processing circuitry configured to implement

a data pre-processing engine providing pre-processed datasets of individuals measured or captured in a past measuring time window, wherein a plurality of data sources are accessed by means of the data pre-processing engine via a data-interface capturing parameter values to each individual of the cohort or portfolio and assigning the parameter values to a corresponding data set associated with an individual, and wherein the data set at least comprises measuring parameter related to demographic characteristics parameters and/or in-force risk-transfer characteristics parameters and/or claims characteristics parameters and/or UW process characteristics parameters and/or extended condition parameters,

wherein the extended condition parameters comprise at least wearable telematics parameters at least comprising log in/out time and/or activities and/or body characteristics data during/between activities and the demographic characteristics parameters comprise at least an occupation class value and/or a height value and/or a weight value and/or a gender value and/or a smoking status value and/or an education indication value and/or a date of birth (DOB) value and/or a race indication value and/or a nationality indication value and/or a marital status value and/or a number of children and/or occupation (sub-) classes level 1 value provided by industry code mapping and/or an issue date and/or an annual income value and/or location value comprising geo-coding data or address data and/or payment frequency and/or payment method indication data,

wherein the data sources of the system at least comprise laboratory/clinical/medical measuring devices, the laboratory and/or clinical and/or medical measuring devices and/or sensors at least comprise measuring devices or sensors measuring platelet distribution width and/or mean platelet volume and/or mean red blood cells volume and/or glucose concentration and/or red blood cells and/or monocytes, lymphocytes and/or mean hemoglobin per red blood cell concentration and/or white blood cells and/or platelets and/or creatinine concentration and/or hemoglobin concentration and/or mean hemoglobin per red blood cell and/or red blood cells distribution width and/or hematocrit, and/or neutrophils,

wherein the processing circuitry is further configured to implement a risk-labelling engine comprising an unsupervised machine-learning (ML) structure for automated clustering the pre-processed datasets of individuals, the risk-labelling engine providing datasets of individuals segmented by clusters by processing the pre-processed datasets of individuals by the unsupervised machine-learning (ML) structure,

wherein the unsupervised machine-learning (ML) structure at least comprise a K-means machine-learning (ML) structure, the K-means machine-learning structure using a vector quantization process partitioning the measured and pre-processed datasets of individuals into a number k of clusters, where each measured and pre-processed dataset of individuals is associated with a cluster having the nearest means to its cluster center or to its cluster centroid serving as a prototype of said cluster,

wherein the ML-based, predictive, digital underwriting system comprises a supervised machine-learning (ML) structure for automated classification and/or validation of datasets of individuals segmented by the clusters, wherein the clusters are automatically classified by a supervised machine-learning structure or by a pattern recognition module based on historically measured and validated datasets, and wherein the learning phase of the supervised machine-learning (ML) structure comprises a feedback learning loop based on classified historical datasets of individuals, and

wherein the ML-based, predictive, digital underwriting system comprises an artificial-intelligence (AI) module with a supervised machine-learning structure trained during a training phase by supervised machine-learning, the AI-module generating in the application phase for each inputted dataset of an individual a measured risk score value as the measured occurrence probability value for the occurrence of one or more predefined medical and/or health and/or life events to the individual assigned to the inputted dataset.

2. The ML-based, predictive, digital underwriting system according to claim 1 , wherein the in-force risk-transfer characteristics parameters comprise at least risk-transfer characteristics parameters and/or contact date and/or time and/or APE (Annual Premium Equivalent) and/or existing UW decision parameters and/or total in-force annualized premium earned (APE) and/or length of relationship and/or last risk-transfer date and/or number of in forced/lapsed risk-transfers.

3. The ML-based, predictive, digital underwriting system according to claim 1 , wherein the existing UW decision parameters comprise at least standard UW decision parameters and/or substandard UW decision parameters and/or rejected UW decision parameters and/or exclusion UW decision parameters.

4. The ML-based, predictive, digital underwriting system according to claim 1 , wherein the claims characteristics parameters comprise at least an incur data value and/or a settle date value and/or a claim type and/or an amount value and/or a frequency and/or disease diagnosis data and/or IDC-10 code (International Statistical Classification of Diseases and Related Health Problems) and/or diagnosis data associated with a medical and/or health and/or life claim and/or impairment code and/or benefit breakdown data.

5. The ML-based, predictive, digital underwriting system according to claim 1 , wherein the UW process characteristics parameters comprise at least a sum value assured and/or a term value of the risk-transfer and/or health characteristics data and/or medical cover parameter and/or past and recent UW decision parameters and/or personal disclosure data.

6. The ML-based, predictive, digital underwriting system according to claim 1 , wherein the extended condition parameters comprise at least agent channel parameters and/or loyalty program parameters and/or bank transaction parameters.

7. The ML-based, predictive, digital underwriting system according to claim 6 , wherein the agent channel parameters at least comprise agency type and/or agency rating and/or agent area characteristics and/or education level and/or education rank and/or tenure parameters and/or persistence parameters.

8. The ML-based, predictive, digital underwriting system according to claim 6 , wherein the loyalty program parameters at least comprise joint data and/or number of activities and/or number reward points earned and/or customer tier parameters.

9. The ML-based, predictive, digital underwriting system according to claim 6 , wherein the bank transaction parameters at least comprise credit card transactions parameters and/or standard industry code of merchant.

10. The ML-based, predictive, digital underwriting system according to claim 1 , wherein the unsupervised machine-learning (ML) structure further at least comprise a mini-batch K-means machine-learning (ML) structure.

11. The ML-based, predictive, digital underwriting system according to claim 10 , wherein, for signal processing, the unsupervised machine-learning (ML) structure is based on a mini-batch K-means structure, the mini-batch K-means structure using small, random, fixed-size batches of the pre-processed dataset of individuals to store in memory, and then collecting with each iteration, a random sample of the pre-processed dataset of individuals used to update the clusters.

12. The ML-based, predictive, digital underwriting system according to claim 11 , wherein the mini-batch K-means or K-means machine-learning (ML) structure is realized based on Python.

13. The ML-based, predictive, digital underwriting system according to claim 1 , wherein for automated classification and/or validation of datasets of individuals, alternatively, the clusters/segments are validated by providing a feedback loop to at least one human expert based on historically measured and validated datasets by means of the data interface of the system.

14. The ML-based, predictive, digital underwriting system according to claim 12 , wherein the historically measured and validated datasets at least comprise data indicating an underwriting decision and/or a claim severity and/or a loss ratio.

15. The ML-based, predictive, digital underwriting system according to claim 1 , wherein the system is calibrated by a calibration module to a predefined calibration threshold, wherein the measured risk score values and/or a measured distribution of the measured risk score values are adjusted to a predefined value range given by the calibration threshold value.

16. The ML-based, predictive, digital underwriting system according to claim 15 , wherein the calibration threshold value represents a risk appetite associated with a risk-transfer system as maximum value of possibly to be covered risks by a risk-transfer system and/or by a specific portfolio comprising a plurality of selected risk-transfers.

17. The ML-based, predictive, digital underwriting system according to claim 15 , wherein a measured performance of an automated risk-transfer system is automatedly adapted by adjusting and/or calibrating the system by measuring a business impact of an error caused by the AI-module to the automated risk-transfer system.

18. The ML-based, predictive, digital underwriting system according to claim 1 , wherein the data sources of the system further at least comprise diagnostic tools and/or medical/clinical databases.

19. A method, implemented by processing circuitry of a machine-learning (ML) based, predictive, digital underwriting system providing an automated parameter-driven predictive underwriting process based on measured probability values associated with individuals of a cohort or portfolio, the individuals being exposed to a probability of occurrence of one or more predefined medical and/or health and/or life events having the probability value with a predefined severity within a future measuring time-window, the method comprising:

implementing a data pre-processing engine providing pre-processed datasets of individuals measured or captured in a past measuring time window, wherein a plurality of data sources are accessed by means of the data pre-processing engine via a data-interface capturing parameter values to each individual of the cohort or portfolio and assigning the parameter values to a corresponding data set associated with an individual, and wherein the data set at least comprises measuring parameter related to demographic characteristics parameters and/or in-force risk-transfer characteristics parameters and/or claims characteristics parameters and/or UW process characteristics parameters and/or extended condition parameters, wherein the extended condition parameters comprise at least wearable telematics parameters at least comprising log in/out time and/or activities and/or body characteristics data during/between activities and the demographic characteristics parameters comprise at least an occupation class value and/or a height value and/or a weight value and/or a gender value and/or a smoking status value and/or an education indication value and/or a date of birth (DOB) value and/or a race indication value and/or a nationality indication value and/or a marital status value and/or a number of children and/or occupation (sub-)classes level 1 value provided by industry code mapping and/or an issue date and/or an annual income value and/or location value comprising geo-coding data or address data and/or payment frequency and/or payment method indication data,

wherein the data sources of the system at least comprise laboratory/clinical/medical measuring devices, the laboratory and/or clinical and/or medical measuring devices and/or sensors at least comprise measuring devices or sensors measuring platelet distribution width and/or mean platelet volume and/or mean red blood cells volume and/or glucose concentration and/or red blood cells and/or monocytes, lymphocytes and/or mean hemoglobin per red blood cell concentration and/or white blood cells and/or platelets and/or creatinine concentration and/or hemoglobin concentration and/or mean hemoglobin per red blood cell and/or red blood cells distribution width and/or hematocrit, and/or neutrophils,

wherein the method further includes implementing a risk-labelling engine comprising an unsupervised machine-learning (ML) structure for automated clustering the pre-processed datasets of individuals, the risk-labelling engine providing datasets of individuals segmented by clusters by processing the pre-processed datasets of individuals by the unsupervised machine-learning (ML) structure,

wherein the unsupervised machine-learning (ML) structure at least comprise a K-means machine-learning (ML) structure, the K-means machine-learning structure using a vector quantization process partitioning the measured and pre-processed datasets of individuals into a number k of clusters, where each measured and pre-processed dataset of individuals is associated with a cluster having the nearest means to its cluster center or to its cluster centroid serving as a prototype of said cluster,

wherein the ML-based, predictive, digital underwriting system comprises a supervised machine-learning (ML) structure for automated classification and/or validation of datasets of individuals segmented by the clusters, wherein the clusters are automatically classified by a supervised machine-learning structure or by a pattern recognition module based on historically measured and validated datasets, and wherein the learning phase of the supervised machine-learning (ML) structure comprises a feedback learning loop based on classified historical datasets of individuals, and

wherein the ML-based, predictive, digital underwriting system comprises an artificial-intelligence (AI) module with a supervised machine-learning structure trained during a training phase by supervised machine-learning, the AI-module generating in the application phase for each inputted dataset of an individual a measured risk score value as the measured occurrence probability value for the occurrence of one or more predefined medical and/or health and/or life events to the individual assigned to the inputted dataset.