International Business Machines Corporation

Workday, Inc.

Eugene Irving Kelton

Brandon Harris

Yi-Hui Ma

Chaz Vollmer

Willie Robert Patten, Jr

Shuyan Lu

John H Walczyk, Iii

Jacob McPherson

Russell Gregory Lambert

Jason Howard Cornpropst

Eliza Salkeld

Shanna Hayes

Arvin Bhatnagar

Jr Patten Willie Robert

Willie Robert Patten

A system, platform, program product, and/or method for generating new composite insight templates that includes: running a machine learning model on a data set to obtain for each of a plurality of entities a risk score and feature-based insights; generating a list of top 'n' features input to the machine learning model that contributes to the risk score for each entity; grouping entities based upon similar features input to the machine learning model that contributes to the risk score for each entity; generating a decision tree for at least one of the group of entities; extracting, from the decision tree generated for the at least one of the group of entities, one or more feature-based insights; generating, by applying subject matter input, a new composite insight based upon the one or more feature-based insights; and adding the new composite insight to insight templates.

System and method to generate insight templates for risk probability analysis

Embodiments of the present invention provide a computer system, a computer program product, and a method that comprises receiving transactional data from at least two users in a plurality of users; determining a pattern within the received transactional data based on a frequency-based domain conversion, wherein the pattern is associated with a determined periodicity; determining a delay within the received transactional data by identifying contextual factors associated with the received transactional data and measuring an amount of time between each identified contextual factors within a plurality of identified contextual factors using a signal processing algorithm; aligning the received transactional data from the at least two users by placing at least two signal forms associated based on the determined delay within the received transactional data at a same point; and generating a line graph depicting the aligned transactional data.

Vector alignment of signal lag

Embodiments of the present invention provide a computer system, a computer program product, and a method that comprises determining a pattern within received data based on a periodicity associated with the received data; in response to a calculated signal score associated with the determined pattern of the received data meeting or exceeding a predetermined threshold transaction amount, tracing at least one location associated with the received data; and dynamically suspending an action associated with an account that generated the received data in response to the traced location not being associated with a historical baseline of the account.

Scaling transactions with signal analysis

An abstraction system for generating a standard customer profile may receive customer data and perform unsupervised learning on the customer data to produce a plurality of clusters of customers with a plurality of features in common, determine that a cluster represents a standard customer and store a plurality of standard customer profiles based on the determined standard customers. The abstraction system may also provide the standard customer profiles to a cognitive system for generating synthetic transaction data based on the standard customer. Generating synthetic transaction data includes selecting a standard customer profile as a goal, simulating a plurality of transactions, comparing the plurality of transactions with the goal, providing feedback, adjusting a policy based on the feedback, repeating until a degree of similarity between the plurality of transactions and the goal is higher than a predefined threshold, and outputting the resulting plurality of transactions as the synthetic transaction data.

System and method for unsupervised abstraction of sensitive data for detection model sharing across entities

In an approach for managing an open feature library, a processor collects feature metadata from a data source, the feature metadata being data related to features for training a machine learning model. A processor builds a feature pipeline for the features based on the collected feature metadata. The feature pipeline may be a set of defined and active features. A processor executes the feature pipeline including computing the features. A processor executes a machine learning model using the computed features. A processor evaluates the features and corresponding model results.

Open feature library management

Embodiments of the present invention provide a computer system a computer program product, and a method that comprises converting received data from a time-based domain to a frequency-based domain using a signal processing algorithm; identifying transactional noise within the converted data by identifying contextual factors based on a determined pattern, wherein the transactional noise is data associated with an identified fraudulent transaction; filtering the identified transactional noise by removing datapoints within the converted data that reaches a predetermined threshold of signal strength using the signal processing algorithm; and generating a line graph depicting removal of the data that is indicative of the identified transactional noise from the converted data.

Removal of transaction noise

A method, computer system, and a computer program product for labeling optimization is provided. The present invention may include receiving a plurality of labeled historical transaction timeline image clusters based on a plurality of historical transaction timeline images clustered using unsupervised machine learning. The present invention may further include training an image recognition model using supervised machine learning based on the received plurality of labeled historical transaction timeline image clusters. The present invention may also include receiving, by the trained image recognition model, a current transaction timeline image. The present invention may further include assigning a corresponding label to the received current transaction timeline image based on matching the received current transaction timeline image to one of the received plurality of labeled historical transaction timeline image clusters.

Labeling optimization through image clustering

Embodiments can provide a computer implemented method for simulating customer data using a reinforcement learning model, including: generating an artificial customer profile by combining randomly selected information from a set of real customer profile data; providing standard customer transaction data representing a group of real customers having similar transaction characteristics as a goal; performing a plurality of iterations to simulate the standard customer transaction data; and combining the artificial customer profile with the simulated customer transaction data to form simulated customer data. In each iteration, the method includes conducting an action including a plurality of simulated transactions; comparing the action with the goal; providing a feedback associated with the action based on a degree of similarity relative to the goal; adjusting a policy based on the feedback.

Intelligent agent to simulate customer data

Computer vision and deep learning techniques are leveraged to detect behavior patterns in transaction histories. A transaction timeline is built for a series of transactions, e.g., financial, and a graphic image is constructed representing the transaction timeline. The graphic image is then matched to a known behavior pattern using a cognitive system. The cognitive system is trained with historical timeline images having associated labels. In one example the graphic image is a bar chart and each financial transaction is represented as a bar in the bar chart having a height proportional to a transaction amount, the bar being located along a time axis of the bar chart according to the transaction date and being color coded according to the transaction type.

Classifying behavior through system-generated timelines and deep learning

A method of preparing a dataset may comprise calculating a pattern relevance for a first field in the dataset and modifying the first field based on the pattern relevance. The method may further comprise detecting a contextual cue in the first field. The method may further comprise retrieving contextual information for a value in the first field and adding that contextual information to the database. Finally, the method may further comprise identifying a numerical scheme for the first field and parsing the first field into a number according to that numerical scheme.

Automated machine-learning dataset preparation

Embodiments can provide a computer implemented method for simulating transaction data using a reinforcement learning model, the method including: generating an artificial customer profile by combining randomly selected information from a set of real customer profile data; providing standard customer transaction data representing a group of real customers having similar transaction characteristics as a goal; conducting, by the intelligent agent, an action including a plurality of simulated transactions; comparing, by the environment, the action with the goal; providing, by the environment, a feedback associated with the action based on a degree of similarity relative to the goal; adjusting, by the policy engine, a policy based on the feedback; the step of conducting an action to the step of adjusting a policy are repeated until the degree of similarity is higher than a first predefined threshold; and combing the artificial customer profile with the action to form simulated customer data.

An approach is provided that automatically computes a number of feature combinations based on a set of subject matter fields that are selected from a set of subject matter fields and further based on qualifiers that correspond to the selected set of subject matter fields. A model in an artificial intelligence (AI) system is then trained using the computed plurality of feature combinations.

Automatic combinatoric feature generation for enhanced machine learning

Embodiments can provide a method for identifying a behavioral pattern from simulated transaction data, including: simulating transaction data using a reinforcement learning model; identifying a behavioral pattern from the simulated transaction data; comparing the behavioral pattern with standard customer transaction data to determine whether the behavioral pattern is present in the standard customer transaction data. If the behavioral pattern is present in the standard customer transaction data, the behavioral pattern is applied in a model implemented on the cognitive system. The step of simulating transaction data further includes: providing standard customer transaction data representing a group of customers having similar transaction characteristics as a goal; and performing a plurality of iterations to simulate the standard customer transaction data, wherein the plurality of iterations is performed until a degree of similarity of simulated customer transaction data relative to the standard customer transaction data is higher than a first predefined threshold.

Identification of behavioral pattern of simulated transaction data

An abstraction system for generating a standard customer profile in a data processing system has a processing device and a memory. The abstraction system may receive customer data from a computing device over a network, perform unsupervised learning on the customer data to produce a plurality of clusters of customers with a plurality of features in common, and determine that a cluster represents a standard customer, and store a plurality of standard customer profiles based on the determined standard customers, wherein the standard customer profiles comprise a plurality of data distributions for the plurality of features in common. The abstraction system also derives additional standard customer profiles by applying a boundary limiter to the customer data. The abstraction system additionally provides the standard customer profiles and the additional standard customer profiles to a cognitive system for generating synthetic transaction data.

System and method for unsupervised abstraction of sensitive data for realistic modeling

Embodiments can provide a computer implemented method for simulating transaction data using a reinforcement learning model including an intelligent agent, a policy engine, and an environment, the method including: providing, by a processor, standard customer transaction data representing a group of customers having similar transaction characteristics as a goal; conducting, by the intelligent agent, an action including a plurality of simulated transactions; comparing, by the environment, the action with the goal; providing a feedback, by the environment, the action based on a degree of similarity relative to the goal; and adjusting, by the policy engine, a policy based on the feedback; the step of conducting an action to the step of adjusting a policy are repeated until the degree of similarity is higher than a first predefined threshold.

Growing community of inventors

Mechanicsburg, PA, United States of America

Eugene Irving Kelton