AI for Credit Scoring in Lending Decisions

Unsecured loan system that allows customers to obtain loans with favorable terms closer to secured loans by using their existing credit accounts as collateral. The system determines interest rates and pre-authorization hold amounts using machine learning models. It analyzes factors like credit scores, account balances, and payment histories to categorize borrowers and calculate confidence scores. This allows optimizing collateral selection and hold amounts. By leveraging existing credit, lenders can offer more options with lower risk compared to traditional unsecured loans.

Source

Evaluating transaction requests received by a computer system using a machine learning algorithm to improve accuracy in granting versus declining transactions. The computer system trains a machine learning model using historical transaction requests. When a new request comes in, the model scores it and compares to a threshold. If above, the request is granted, if below, declined. This improves accuracy compared to just using a fixed threshold as the model learns from prior requests. However, wrongly rejecting or granting transactions can degrade performance. To mitigate this, the model is retrained periodically using a subset of recent requests to adapt to changing conditions. This allows updating the model without having to retrain from scratch.

Source

Automated system for building customized machine learning models tailored to specific applications like credit underwriting, using a sampling process to speed up model development. The system prompts users to input business-specific data like credit bureau, region, loan type, etc. It then samples a subset of national credit data matching those criteria. This lookalike sample is used to train the customized model instead of the full lender dataset. This allows faster model development by reducing data gathering. The system also presents a user interface to test the customized models using real data.

Source

Dynamic modeling system for credit scoring that uses machine learning and real-time credit value modeling to provide customized credit options and conditions for users. The system generates user interface elements that dynamically present account-specific credit values and conditions based on user activity and risk analysis. It uses an activity machine learning model to calculate a risk score from user activity data, then a credit value model to determine dynamic credit ranges and conditions for selected credit values. This allows flexible, efficient, and accurate presentation of customized credit options and conditions tailored to each user's activity level and risk profile.

Source

Intelligent lending platform using microservices, blockchain, and AI to enable adaptive and automated lending transactions. The platform has services for data collection, blockchain, smart contracts, and user interfaces to handle lending activities and events. It leverages multi-modal data collection from IoT, crowdsourcing, and social networks to monitor collateral and loan conditions. Smart contracts automate loan terms based on monitored data. AI optimizes loan terms and conditions. The platform provides adaptive lending solutions across the loan lifecycle.

Source

Intelligent lending platform that uses machine learning to analyze denied loan applications, identify the main reasons for denial, and provide customized and actionable explanations back to the applicant. The platform trains a complex machine learning model on historical loan data to accurately predict why a particular loan application was denied. It then extracts the most significant reason(s) and presents them in a clear and understandable way to the applicant. This allows the applicant to see exactly why they were denied and provides specific actions they can take to improve their chances for future loans. The platform monitors interactions and adjusts lending decisions based on actions taken on the recommendations.

Source

Credit scoring model that improves accuracy and robustness by leveraging social network data. The model calculates a user's credit score using their personal score as well as scores of other users they have social connections with. This expands coverage and reduces errors compared to just using personal data. The model is trained iteratively with user scores, relative scores based on connections, and default annotations.

Source

Estimating creditworthiness of loan or credit applicants who have insufficient domestic credit history by analyzing their email messages. If a credit request has insufficient domestic credit data, the credit decision platform obtains access to the applicant's email account and uses machine learning to identify relevant messages. It analyzes the content using natural language processing to generate non-domestic credit data. Metrics from this data are used to estimate creditworthiness.

Source

Automated credit building system that uses machine learning to recommend actions to users for improving their credit scores. The system collects user data, activity, and credit scores, assesses potential actions, and provides personalized recommendations to reach credit goals. It also rewards users for meeting thresholds and leverages third-party apps to suggest income sources based on user needs.

Source

Using machine learning models to accurately predict the likelihood of future events that impact product suitability for specific consumers, like loan defaults. The method involves iteratively training a risk-evaluation model using historical data to predict the likelihood of future events like payment defaults. This prediction is then used to evaluate product suitability and make informed decisions about providing products like loans. By leveraging machine learning, the model can account for variations in available data length and provide more accurate predictions compared to traditional rules-based approaches.

Source

A system for training fairer predictive models in domains like credit risk, drug evaluation, and college admissions. The system uses adversarial training techniques to improve fairness while maintaining accuracy. It involves training a primary model to predict an outcome, then training a secondary model to predict the sensitive attribute based on the primary model's output. The primary model is then retrained to maximize the secondary model's accuracy while minimizing its own. This forces the primary model to learn features that are invariant to the sensitive attribute, making it more fair. The system also provides tools for selecting and justifying the best fairness-accuracy tradeoff.

Source

Determining financial risk based on self-supervised natural language extraction from unstructured data sets like long form financial narratives. The method involves converting unstructured financial narratives into condensed financial risk narratives using self-supervised natural language processing. A tokenization library is determined for the unstructured data and used to generate the condensed narratives. Security scores indicative of financial risk are calculated for the condensed narratives. If a score exceeds a threshold, security actions are executed. This allows autonomous risk assessment from unstructured financial data without manual summarization.

Source

Predicting financial well-being of individuals without requiring them to complete surveys, using machine learning models trained on personal data. The models learn to correlate personal data sets of users with their assessment scores. To predict a user's score, the model matches their personal data to that of other users. If a difference between predicted and test scores exceeds a threshold, actions are taken like sending communications or changing account settings. The models use unsupervised learning on user data sets without labels.

Source

A cloud-based digital platform for analyzing and managing risk in financial transactions like loans. The platform ingests diverse types of data from various sources, extracts relevant information, validates accuracy, combines elements, analyzes to reduce credit risk, enables automated workflows, provides customizable reporting, and monitors loan life cycle. It aims to improve loan processing efficiency and risk control through enhanced data ingestion, accuracy checks, data quality control, integrated analysis, automated processing, and reporting.

Source

Aggregating a user's transaction data from multiple sources to determine a credit metric for credit decisions. The method involves receiving aggregated transaction data from various third-party sources where the user has accounts, analyzing it using machine learning to determine a credit metric, and providing the credit metric to interested parties. This provides a more complete view of the user's financial activity for credit evaluation than just relying on a credit bureau score.

Source

Multi-dimensional creditworthiness evaluation using machine learning to provide more accurate and timely credit risk assessment compared to traditional credit scoring methods. The method involves analyzing a user's financial transactions to quantify their creditworthiness along multiple dimensions, such as financial behavior, priority expenses, credit exclusivity, financial discipline, and red-flag events. Machine learning models are trained using this multi-dimensional data to predict creditworthiness for new users. This provides a more comprehensive and dynamic assessment of creditworthiness that addresses shortcomings of traditional credit scores like lag time and limited history for thin-file users.

Source

Improving accuracy of automated predictions like credit scoring by identifying clusters of similarly situated underperforming entities (UEs) and boundary performing entities (PEs) around them. These clusters reveal nuanced patterns and differences between UEs and PEs within categories. By fine-granularly adjusting prediction models based on cluster insights, it allows tailoring predictions for individual entities rather than broadly penalizing categories. It involves calculating deep-credit scores, finding closest PEs for UEs, forming clusters, and tracking distances to evaluate servicing effectiveness.

Source

Predicting credit scores for people who don't have traditional credit histories using their driving behavior data. The method involves training a credit score prediction model using historical user data, vehicle telematics data, and credit scores. The model is then used to predict credit scores for new users based on their driving behavior and other data.

Source

A credit risk assessment method using deep learning to analyze text and financial data together for more accurate credit risk modeling. The method involves a two-stage neural network to process documents and company data separately. The document model understands relationships between words and phrases in unstructured text to generate scores indicating likelihood of financial events. The company model aggregates document scores with financial data to produce default probability sequences. This allows leveraging the semantic context of text to improve credit risk modeling beyond just financial ratios.

Source

Using AI agents to generate credit attributes for customers based on their credit history and other factors, and using those attributes to recommend invoice deferrals and amounts. The AI agents analyze subsets of the credit data to generate individual credit attributes for each customer. These attributes are then used to determine appropriate deferral terms for pending invoices. The AI-generated credit attributes provide a more objective and consistent way to assess deferral requests compared to manual judgments by agents.

Source

Expanding user coverage of credit services on application platforms by training credit scoring models specifically for users with unknown credit history. The method involves identifying a subset of users with unknown credit scores but higher predicted profits. Samples from this subset are used to train a scoring model. Then, the trained model is applied to all users to score their creditworthiness. This allows accurate credit scoring for previously unknown users, expanding the user base that can access credit services through the platform.

Source

Data exploration analysis based covenants categorization and recommendation system for loan origination that leverages historical loan data and machine learning to recommend covenants for new loans. The system trains binary and classification machine learning models using historical loan data to categorize covenants into two categories. It then iteratively trains intermediate models for the second category until the number of predicted covenants matches a threshold. This multi-step training process improves the accuracy of covenant recommendations for new loans. The system provides a real-time covenant recommendation engine using machine learning that leverages historical loan data to recommend covenants for new loans based on customer, industry, and loan details.

Source

Neural network technique to provide embeddings of transaction data for tasks like anomaly detection, fraud analysis, credit decision making, etc. The technique involves training a neural network with transaction data to learn low-dimensional dense representations, called embeddings, of entities in the network graph. The embeddings capture relationships between entities like consumers and merchants. After training, new transactions can be analyzed by encoding them into embeddings and comparing similarities to known embeddings for tasks like anomaly detection or credit decision making.

Source

Estimating stability of dynamic datasets over time to predict financial stability of customers based on observable attributes. The method involves capturing dynamic data like financial transactions into vectors, mapping them to directionally similar template states, generating features from the sequence of template states, and applying a classification algorithm to identify trends in the underlying dynamic data. This allows estimating the stability of a customer's finances over time using observable data rather than just predicting creditworthiness at a single point in time.

Source

Efficient decision-making system using supervised learning for credit scoring and other applications where a model needs to predict a performance variable like creditworthiness. The system trains a decision engine using labeled data and then uses it to calculate decision scores for new entities based on their attributes. The engine breaks down the decision into linearly decomposable components corresponding to each attribute. This allows customizable compliance checks and explanations of how each factor contributes to the overall score.

Source

Using artificial intelligence to improve loan approval decisions and expand lending opportunities to non-traditional applicants with limited credit and employment history. The AI agent analyzes applicant data beyond just credit scores and work history. It retrieves ratings from customer review sites to assess the applicant's reputation and character. This additional input, along with financial and employment info, is used to determine loan recommendations for non-traditional applicants. The AI also detects loan application fraud using video and audio analysis during virtual interviews.

Source

Exchanging user data through federated machine learning to improve loan underwriting. The method involves training a shared neural network model across multiple entities like banks and merchants using their customer data. This federated model is used to generate risk scores for loan applicants based on their existing financial history. These scores are then shared with lenders to help them assess loan applications. The federated training allows leveraging of pooled customer data from multiple sources to build a more comprehensive creditworthiness model.

Source

Intelligently matching and validating data for loan underwriting using predictive modeling techniques. The system receives loan application data and compares it to guidelines and external sources to validate and verify key elements. It leverages blockchain-like ledgers for immutable underwriting history. Bayesian networks predict loan characteristics based on prior data. Adaptive logic determines linguistic distances between application data and stored/retrieved data to verify accuracy. The system automates underwriting using machine learning, deep learning, and blockchain technologies to optimize loan processing efficiency and reduce errors.

Source

Credit decisioning and management using graph neural networks (GNNs) that can handle the high dimensionality and sparsity of transaction networks. The method involves training GNNs on graph representations of credit relationships between borrowers and lenders. The graphs have nodes representing borrowers or lenders with embedding vectors of features. The GNNs propagate messages between nodes based on their relationships, like loan amounts, to update node embeddings. This allows learning credit risk factors from the sparse network structure. The GNNs can be used for credit decisioning and line management by analyzing a borrower's graph neighborhood.

Source

A system for faster and more efficient calculation of credit risk for multiple companies using machine learning. It involves training predictive models using historical distance-to-default data for some companies, and then using those models to forecast changes in distance-to-default for other companies without the iterative Merton model calculation. This allows faster and less computationally intensive assessment of credit risk for large sets of companies compared to individual Merton model calculations.

Source

Intelligent credit limit adjustment for service providers like credit card companies using machine learning models to predict and forecast entity behavior. The system takes initial data as input and uses trained ML models to predict factors like risk assessment and forecasted global balance. This allows dynamic credit limit adjustment based on real-time data rather than static rules. It addresses limitations of conventional systems that use outdated data for credit decisions.

Source

Using machine learning models to generate more accurate and efficient predictions for approving financial service requests (FSRs) like loan applications. The system involves a prediction manager that analyzes FSRs and historical data to generate prediction inputs. It also extracts comments from FSRs and agents to generate a request vector representing the authenticity of the request. The manager applies machine learning models to the prediction inputs and request vector to generate initial predictions. It provides the predictions to approvers and obtains comments to further refine the predictions using chained request vectors. This chained analysis improves subsequent predictions by capturing the authenticity of the request and approver comments.

Source

Mitigating algorithmic bias in credit approval decisions using machine learning models. The method involves generating an inferred protected class dataset based on applicant profile data like name and address. This inferred dataset is used along with the training dataset to determine fairness metrics for the credit approval decisions. The credit approval model is then adjusted to increase these fairness metrics and mitigate bias. Techniques like removing discriminatory features and checking disparate impact are used to mitigate bias.

Source

Using heuristic algorithms to assess credit risk by processing natural language inputs and unstructured data sets to reduce credit risk of customer transactions. The method involves retrieving transaction data, executing a heuristic algorithm using the data to generate credit scores for new transactions, and continually training and refining the algorithm based on feedback. This allows improving credit risk assessment over time by leveraging existing transaction history and context.

Source

Explainable artificial intelligence (AI) system for financial decision making that provides transparent and auditable explanations for AI-based financial transaction decisions. The system uses a neural network model for decision making, but also has an explainability module that analyzes the neural network weights to determine feature importance and node impact. This allows tracing back through the network to explain how each input feature contributes to the final decision. The explainability module can also identify biases and imbalances in the data. The explained decisions are presented to users, along with options to override the AI if desired. The system also allows defining functional guidelines to mitigate errors.

Source

System to extract and manage obligations for delayed payment products like buy now pay later (BNPL) using AI and machine learning to analyze user history and make approval decisions. The system extracts user's current and historic interval obligations, like BNPL, and analyzes them using a pre-endorsement matrix. When a user initiates a new BNPL, the system can modify terms in real-time based on historic obligation satisfaction. It also uses obligation mapping to determine approval for future BNPL based on user history. This allows monitoring and management of fragmented BNPL providers to mitigate exposure for issuing entities.

Source

Using recurrent neural networks (RNNs) to predict future financial health of entities like businesses and individuals, and providing recommendations to improve financial outcomes. The RNNs are trained on historical financial data to predict future equity statuses. If a prediction indicates financial distress, the system adjusts the input features to train the RNN to instead predict normal financial health. This allows iterative refinement to find actions that prevent distress. The system provides both the distress prediction and recommended actions to users.

Source

A lending transaction enablement platform using data-integrated microservices like blockchain, AI, and IoT to improve lending efficiency, reduce risk, and enable adaptive intelligence. It leverages services like data collection, monitoring, smart contracts, crowdsourcing, and automation to enhance lending processes like loan negotiation, underwriting, marketing, compliance, rating, and debt management. The platform can leverage data like IoT sensor readings, social media, and crowdsourced inputs to dynamically adjust loan terms based on factors like regulatory requirements, market conditions, collateral value, and borrower reliability. It also provides features like automated loan restructuring, smart contract-based loan execution, and automated compliance monitoring.

Source

Classifying users to determine if they will apply for a microloan using a machine learning model trained on categorized user data. The method involves obtaining user device data and external data sources, categorizing it, preprocessing features, selecting important features, balancing the dataset, training the model on classification, and using the model to predict loan application likelihood. This allows accurately identifying potential loan applicants with low resource consumption.

Source

Automated underwriting and processing loans using machine learning to quickly approve loans without human intervention while complying with regulations. The system receives loan applications, collects external data, pre-processes it using ML models, does automated feature engineering, determines business objectives, creates ML models, and creates an ensemble model. It also maps adverse action notices to categories. This allows fully automated loan processing that maximizes lender valuation using ML models trained on pre-processed external data.

Source

Credit risk control using graph neural networks trained on enterprise relationship networks. The method involves obtaining a graphical structure model trained on labeled samples to identify high-risk nodes in an enterprise network. This model calculates embedding vectors for nodes based on their original features and relationships. To assess credit risk of new nodes, their embedding vectors are calculated using the trained model, and risk is determined based on the vector values.

Source

Transaction data processing method for financial analysis using graph convolutional networks (GCNs) and transformers to extract spatial-temporal features from transaction graphs. The method involves obtaining transaction data for an account over multiple time windows, extracting spatial features using GCNs and temporal features using transformers, and generating a feature representation for the account based on the combined spatial-temporal information. This representation can be used for downstream analysis tasks like credit risk modeling, fraud detection, and money laundering detection.

Source

Automating loan decision making by training a machine learning model to classify loan applications based on underwriter comments. The method involves extracting factors from unstructured underwriter comments using text mining, feeding those factors into a trained ML model, and using the model's output classification instead of manual decision making. This allows automated loan approval/rejection based on the same factors as human underwriters. The ML model is trained on labeled comments and decisions from a corpus of underwriter actions.

Source

Optimizing credit risk segmentation to accurately differentiate loan default rates. The technique uses mixed integer programming to group credit score bins into segments with maximized linear slopes of default rates. This provides segments with monotonic default rates, improving segmentation accuracy compared to manual methods.

Source

Predicting future credit defaults using adaptively trained AI models and normalized data. The method involves generating an input dataset with normalized features from customer interactions during an initial delinquency event. An AI model trained on this input predicts the likelihood of a future default event after the delinquency event. This allows identifying customers at high risk of defaulting beyond a certain duration of delinquency. The AI model is adaptively trained using gradient boosted decision trees in a distributed computing environment.

Source

Identifying principal adverse action factors for credit denials using machine learning credit risk models. The method involves calculating replacement scores for each characteristic when the credit request is denied, replacing the applicant's actual score for that characteristic with an anchor value. These replacement scores are then ranked to determine the principal adverse action factors. This allows identifying factors that significantly impacted the denial when using complex machine learning models that are less transparent than scorecards.

Source

Predicting targeted, agency-specific recovery events for delinquent loans using adaptively trained artificial intelligence processes. The method involves training machine learning models to predict recovery rates for each collection agency based on features like customer profiles, transaction data, delinquency history, etc. This allows assigning delinquent loans to the agency with the predicted highest recovery rate. The models are trained using historical data and validated on separate time intervals.

Source

Monitoring user transactions in systems like financial services to more accurately identify and respond to high-risk users. It uses a machine learning clustering technique to group users based on their observed transaction activity features. Users are assigned to clusters based on similarity in transaction patterns. Risks are assessed for each cluster based on feature averages. Users are monitored based on cluster risk. Changes in user features within clusters indicate potential risk. Users moving between clusters or segments are flagged. Seasonal adjustment accounts for activity variations.

Source

Selectively using a heuristic model or a statistical model to analyze relationship attributes of a borrower to determine whether to approve or deny a lending-product request, in a system that determines a probability of a borrower repaying a loan over a predetermined time, and avoiding being charged off. The system uses a Statistical Risk Management (SRM) approach to evaluate borrower risk beyond just creditworthiness checks. It analyzes relationship attributes to predict loan repayment and default likelihoods. This allows more nuanced decision making for small, short-term loans where traditional creditworthiness checks are impractical.

Source

Using machine learning to assess creditworthiness for loans to consumers for purchasing products and services from merchants. The method involves training a machine learning model using financial data from consumers, then analyzing a consumer's parameters along with merchant-specified requirements like costs and reserves. This enables flexible loan approvals where merchants can adjust terms like reserves to increase approval odds.

Source