Machine learning (ML)-based system and method for predicting financial transaction patterns

1 . A machine learning (ML)-based computing system for predicting financial payment transaction patterns in a computing environment, the ML-based computing system comprising:

one or more hardware processors; and

a memory coupled to the one or more hardware processors, wherein the memory comprises a plurality of modules in the form of programmable instructions executable by the one or more hardware processors, and wherein the plurality of modules comprises:

a data receiver module configured to receive invoice data of at least one of one or more customers and one or more vendors from an external database, wherein the invoice data comprises a set of closed invoices and a set of open invoices along with a set of data attributes;

a granularity level instance creation module configured to:

extract a set of granularity levels from the received invoice data based on data density, volume, and the set of data attributes, wherein the set of granularity levels comprises at least one of company code, customer number, business unit level, invoice types, and payment terms; and

generate a set of granularity level instances based on the extracted set of granularity levels, wherein each of the set of granularity level instances corresponds to a basic logical unit for which payment transaction pattern is to be determined, wherein the payment transaction pattern is associated to a payment date;

a payment frequency computation module configured to:

compute a payment frequency bucket feature for each of the set of granularity level instances, wherein the payment frequency bucket features corresponds to one of: a weekly payer, alternative weekly payer, a monthly payer, a bi-monthly, a quarterly payer, a half yearly payer and an annual payer; and

assign the set of closed invoices to a set of clusters based on the determined payment frequency bucket features and the data density, wherein each of the set of clusters comprises closed invoices based on the set of payment frequency bucket features and the data density corresponding to each of the set of granularity level instances;

a payment pattern generation module configured to dynamically generate a set of date-shift-payment pattern features by shifting a set of invoice level date attributes comprising at least one of: discount date, due date, clearing date and invoice date to a set of reference points on a calendar, based on the determined set of payment frequency bucket features assigned to the set of clusters, wherein each of the set of date-shift-payment pattern features indicates a date on which payment transaction is set to occur for each of the set of open invoices,

a payment pattern selection module configured to determine an optimal pattern from among the generated set of date-shift-payment pattern features with a highest probability of adherence for each granularity level instance using a customer pattern adherence process, wherein the payment pattern selection module is configured to:

determine an adherence evaluation window comprising a lower bound value and an upper bound value;

configure a specific lookback period for each of the set of date-shift-payment pattern features, wherein the specific lookback period is calculated as a multiple of a time period of a cycle of the respective date-shift-payment pattern feature;

filter the set of closed invoices that cleared within the configured specific lookback period;

calculate a gap between an actual clearing date and a pattern date predicted by the date-shift-payment pattern feature for the filtered closed invoices;

compute a revenue concentration level within the determined adherence evaluation window by calculating a cumulative sum of the calculated gaps based on revenue percentage; and

select the optimal pattern with a maximum revenue concentration level in the adherence evaluation window for each granularity level instance;

a data prediction module configured to predict a financial transaction payment pattern for the set of open invoices based on the optimal pattern, the set of payment pattern features and the set of granularity level instances by using a financial-transaction-pattern-prediction-based ML model, wherein the financial transaction payment pattern comprises an estimated date at which the one or more customers and the one or more vendors complete a financial transaction to clear the set of open invoices, wherein the financial-transaction-pattern-prediction based ML model is a regression-based ML model, wherein the regression-based ML model is configured to:

pre-process input data comprising the optimal pattern, the set of payment pattern features, and the set of granularity level instances to ensure a format suitable for input into the regression-based ML model;

train a set of decision trees using the pre-processed input data to iteratively correct prediction errors for the financial transaction payment pattern;

apply at least one of: regularization to prevent overfitting, and encoding of categorical features to eliminate need for extensive pre-processing; and

output a weighted average of the predictions from all the set of decision trees;

wherein the ML-based computing system is configured to:

dynamically monitor prediction performance of the regression-based ML model in real time;

identify instances where the prediction performance decreases below a predetermined threshold;

automatically generating new training data based on the identified instances; and

retrain the regression-based ML model using newly generated training data; and

a data output module configured to output the predicted financial transaction payment pattern on an user interface screen of one or more electronic devices associated with a user.

2 . The ML-based computing system of claim 1 , wherein the invoice data comprises a set of closed invoices and a set of open invoices, wherein the invoice data comprises customer number, payment terms, invoice date, baseline date, due date, discount due date, clearing date, posting date, invoice type, business unit within a company, invoice amount, discount amount percentage and payment block status, and wherein the set of open invoices corresponds to the invoices associated with payment transactions which are yet to occur, and wherein the set of closed invoices corresponds to the invoices associated with payment transactions previously occurred.

3 . The ML-based computing system of claim 1 , wherein in determining the optimal pattern from among the generated set of date-shift-payment pattern features with the highest probability of adherence for each granularity level instance using the customer pattern adherence process, the payment pattern selection module is configured to:

obtain the historical invoice data comprising the payment frequency bucket feature based clustered training data and the set of date-shift-payment pattern features;

estimate a cumulative error distribution value of historical errors for each of the set of date-shift-payment patterns corresponding to each of the set of granularity level instances;

estimate a probability value of adherence for each of the set of date-shift-payment pattern features by comparing each of the set of date-shift-payment pattern features with an actual clearing date based on the estimated cumulative error distribution value of historical errors; and

determine the optimal pattern from the set of date-shift-payment pattern features, wherein the optimal pattern corresponds to a pattern with highest probability of adherence for each of the set of granularity level instances.

4 . The ML-based computing system of claim 3 , wherein in estimating the cumulative error distribution value of historical errors for each of the set of date-shift-payment pattern features corresponding to each of the set of granularity level instances, the payment pattern selection module is configured to:

determine an adherence evaluation window for a particular customer, wherein the adherence evaluation window comprises a lower bound value and an upper bound value;

determine a lookback period for each of the set of date-shift-payment pattern features, wherein the lookback period is a multiple of a time period of each of the set of date-shift-pattern-features; and

determine the cumulative error distribution for each of the set of date-shift-payment pattern features per granularity level instances, wherein the cumulative error distribution value corresponds to the revenue concentration levels in the determined adherence evaluation window.

5 . The ML based computing system of claim 4 , wherein estimating the cumulative error distribution value of the historical errors based on the determined lookback pattern mapping value, the payment pattern selection module is configured to:

filter the set of closed invoices based on lookback period of the lookback pattern mapping value, wherein the set of closed invoices cleared within the lookback period are filtered;

calculate a gap between clearing date and a pattern date for each of the pattern in the determined lookback pattern mapping value;

calculate a sum amount and a cumulative sum amount for each calculated gap and the granularity level instance;

estimate the cumulative error distribution value of historical errors by calculating the revenue percentage at each stage respective to historical revenue of the granularity level instance in the lookback period; and

calculate the revenue concentration levels in the determined adherence evaluation window for each of the customers in the estimated cumulative error distribution value of historical errors.

6 . The ML based computing system of claim 3 , wherein in determining the optimal pattern from the set of date-shift-payment pattern features, the payment pattern selection module is configured to:

select the optimal pattern with maximum revenue concentration level in the determined adherence evaluation window for a particular granularity level instances based on the determined lookback pattern mapping value.

7 . The ML based computing system of claim 1 , wherein predicting the financial transaction payment pattern for the set of open invoices based on the optimal pattern, the set of payment pattern features and the set of granularity level instances by using the financial-transaction-pattern-prediction-based ML model, the data prediction module is configured to:

apply the determined optimal pattern, the set of payment pattern features and the set of granularity level instances onto the financial-transaction-pattern-prediction-based ML model;

map each of the determined optimal pattern, the set of payment pattern features and the set of granularity level instances with the set of open invoices; and

predict the financial transaction payment pattern for the set of open invoices based on the mapping.

8 . A machine learning (ML)-based method for predicting financial transaction patterns in a computing environment, the ML-based method comprising:

receiving, by a processor, invoice data of at least one of one or more customers and one or more vendors from an external database, wherein the invoice data comprises a set of closed invoices and a set of open invoices along with a set of data attributes;

extracting, by the processor, a set of granularity levels from the received invoice data based on data density, volume, and the set of data attributes, wherein the set of granularity levels comprises at least one of company code, customer number, business unit level, invoice types, and payment terms; and

generating, by the processor, a set of granularity level instances based on the extracted set of granularity levels, and wherein each of the set of granularity level instances corresponds to a basic logical unit for which payment transaction pattern is to be determined, wherein the payment transaction pattern is associated to a payment date;

computing, by the processor, a payment frequency bucket feature for each of the set of granularity level instance, wherein the payment frequency bucket features corresponds to one of: a weekly payer, alternative weekly payer, a monthly payer, a bi-monthly, a quarterly payer, a half yearly payer and an annual payer;

assigning, by the processor, the set of closed invoices to a set of clusters based on the determined payment frequency bucket features and the data density, wherein each of the set of clusters comprises closed invoices based on the set of payment frequency bucket features and the data density corresponding to each of the set of granularity level instances;

dynamically generating, by the processor, a set of date-shift-payment pattern features based on the determined set of payment frequency bucket features assigned to the set of clusters by shifting a set of invoice level date attributes comprising at least one of: discount date, due date, clearing date and invoice date to a set of reference points on a calendar, wherein each of the set of date-shift-payment pattern features indicates a date on which payment transaction is set to occur for each of the set of open invoices;

determining, by the processor, an optimal pattern from among the generated set of date-shift-payment pattern features with a highest probability of adherence for each granularity level instance using a customer pattern adherence process, wherein determining an optimal pattern from among the generated set of date-shift-payment pattern features, comprises:

determining, by the processor, an adherence evaluation window comprising a lower bound value and an upper bound value;

configuring, by the processor, a specific lookback period for each of the set of date-shift-payment pattern features, wherein the specific lookback period is calculated as a multiple of a time period of a cycle of the respective date-shift-payment pattern feature;

filtering, by the processor, the set of closed invoices that cleared within the configured specific lookback period;

calculate a gap between an actual clearing date and a pattern date predicted by the date-shift-payment pattern feature for the filtered closed invoices;

computing, by the processor, a revenue concentration level within the determined adherence evaluation window by calculating a cumulative sum of the calculated gaps based on revenue percentage; and

selecting, by the processor, the optimal pattern with a maximum revenue concentration level in the adherence evaluation window for each granularity level instance;

predicting, by the processor, a financial transaction payment pattern for the set of open invoices based on the optimal pattern, the set of payment pattern features and the set of granularity level instances by using a financial-transaction-pattern-prediction-based ML model, wherein the financial transaction payment pattern comprises an estimated date at which the one or more customers and the one or more vendors complete a financial transaction to clear the set of open invoices, wherein the financial-transaction-pattern-prediction based ML model uses a regression-based ML model;

pre-processing, by the regression-based ML model, input data comprising the optimal pattern, the set of payment pattern features, and the set of granularity level instances to ensure a format suitable for input into the regression-based ML model;

training, by the regression-based ML model, a set of decision trees using the pre-processed input data to iteratively correct prediction errors for the financial transaction payment pattern;

applying, by the regression-based ML model, at least one of: regularization to prevent overfitting, and encoding of categorical features to eliminate need for extensive pre-processing;

outputting, by the regression-based ML model, a weighted average of the predictions from all the set of decision trees;

dynamically monitoring, by the processor, prediction performance of the regression-based ML model in real time;

identifying, by the processor, instances where the prediction performance decreases below a predetermined threshold;

automatically, by the processor, generating new training data based on the identified instances;

retraining, by the processor, the regression-based ML model using newly generated training data; and

outputting, by the processor, the predicted financial transaction payment pattern on the user interface screen of one or more electronic devices associated with an user.

9 . The ML-based method of claim 8 , wherein the invoice data comprises a set of closed invoices and a set of open invoices, wherein the invoice data comprises customer number, payment terms, invoice date, baseline date, due date, discount due date, clearing date, posting date, invoice type, business unit within a company, invoice amount, discount amount percentage and payment block status, and wherein the set of open invoices corresponds to the invoices associated with payment transactions which are yet to occur, and wherein the set of closed invoices corresponds to the invoices associated with payment transactions previously occurred.

10 . The ML-based method of claim 8 , wherein determining the optimal pattern from among the generated set of date-shift-payment pattern features with the highest probability of adherence for each granularity level instance using the customer pattern adherence process comprises:

obtaining historical invoice data comprising the payment frequency bucket feature based clustered training data and the set of date-shift-payment pattern features;

estimating a cumulative error distribution value of historical errors for each of the set of date-shift-payment patterns corresponding to each of the set of granularity level instances;

estimating a probability value of adherence for each of the set of date-shift-payment pattern features by comparing each of the set of date-shift-payment pattern features with an actual clearing date based on the estimated cumulative error distribution value of historical errors; and

determining the optimal pattern from the set of date-shift-payment pattern, wherein the optimal pattern corresponds to a pattern with highest probability of adherence for each of the set of granularity level instances.

11 . The ML-based method of claim 10 , wherein estimating the cumulative error distribution value of the historical errors for each of the set of date-shift-payment pattern features corresponding to each of the set of granularity level instances comprises:

determining an adherence evaluation window for a particular customer, wherein the adherence evaluation window comprises a lower bound value and an upper bound value;

determining a lookback period for each of the set of date-shift-payment pattern features, wherein the lookback period is a multiple of a time period of each of the set of date-shift-pattern-features; and

determining the cumulative error distribution for each date-shift-payment pattern-feature per granularity level instances, wherein the cumulative error distribution value corresponds to the revenue concentration levels in the determined adherence evaluation window.

12 . The ML based method of claim 11 , wherein determining the optimal pattern from the set of date-shift-payment pattern comprises:

selecting the optimal pattern with maximum revenue concentration level in the determined adherence evaluation window for a particular granularity level instances based on the determined lookback pattern mapping value.

13 . The ML based method of claim 11 , wherein estimating the cumulative error distribution value of the historical errors based on the determined lookback pattern mapping value comprises:

filtering the set of closed invoices based on lookback period of the lookback pattern mapping value, wherein the set of closed invoices cleared within the lookback period are filtered;

calculating a gap between clearing date and a pattern date for each of the pattern in the determined lookback pattern mapping value;

calculating a sum amount and a cumulative sum amount for each calculated gap and the granularity level instance;

estimating the cumulative error distribution value of historical errors by calculating the revenue percentage at each stage respective to historical revenue of the granularity level instance in the lookback period; and

calculating the revenue concentration levels in the determined adherence evaluation window for each of the customers in the estimated cumulative error distribution value of historical errors.

14 . The ML based method of claim 8 , wherein predicting the financial transaction payment pattern for the set of open invoices based on the optimal pattern, the set of payment pattern features and the set of granularity level instances by using the financial-transaction-pattern-prediction-based ML model comprises:

applying the determined optimal pattern, the set of payment pattern features and the set of granularity level instances onto the financial-transaction-pattern-prediction-based ML model;

mapping each of the determined optimal pattern, the set of payment pattern features and the set of granularity level instances with the set of open invoices; and

predicting the financial transaction payment pattern for the set of open invoices based on the mapping.

15 . The ML-based method of claim 8 , wherein the ML-based method further comprises:

monitoring the performance of the ML-based computing system in real-time;

identifying instances wherein the performance of the ML-based computing system decreases below a predetermined threshold;

generating a new training data based on the identified instances; and retraining the financial-transaction-pattern-prediction-based ML model using new training data.

16 . A non-transitory computer-readable storage medium having instructions stored therein that when executed by one or more hardware processors, cause the one or more hardware processors to execute operations of:

receiving invoice data of at least one of one or more customers and one or more vendors from an external database, wherein the invoice data comprises a set of closed invoices and a set of open invoices along with a set of data attributes;

extracting a set of granularity levels from the received invoice data based on data density, volume, and the set of data attributes, wherein the set of granularity levels comprises at least one of company code, customer number, business unit level, invoice types, and payment terms; and

generating a set of granularity level instances based on the extracted set of granularity levels, and wherein each of the set of granularity level instances corresponds to a basic logical unit for which payment transaction pattern is to be determined, wherein the payment transaction pattern is associated to a payment date;

computing a payment frequency bucket feature for each of the set of granularity level instance, wherein the payment frequency bucket features corresponds to one of: a weekly payer, alternative weekly payer, a monthly payer, a bi-monthly, a quarterly payer, a half yearly payer and an annual payer;

assigning, by the processor, the set of closed invoices to a set of clusters based on the determined payment frequency bucket features and the data density, wherein each of the set of clusters comprises closed invoices based on the set of payment frequency bucket features and the data density corresponding to each of the set of granularity level instances;

dynamically generating, by the processor, a set of date-shift-payment pattern features by shifting a set of invoice level date attributes comprising at least one of: discount date, due date, clearing date and invoice date to a set of reference points on a calendar, based on the determined set of payment frequency bucket features assigned to the set of clusters, wherein each of the set of date-shift-payment pattern features indicates a date on which payment transaction is set to occur for each of the set of open invoices;

determining an optimal pattern from among the generated set of date-shift-payment pattern features with a highest probability of adherence for each granularity level instance using a customer pattern adherence process, wherein determining an optimal pattern from among the generated set of date-shift-payment pattern features, comprises:

determining an adherence evaluation window comprising a lower bound value and an upper bound value;

configuring a specific lookback period for each of the set of date-shift-payment pattern features, wherein the specific lookback period is calculated as a multiple of a time period of a cycle of the respective date-shift-payment pattern feature;

filtering the set of closed invoices that cleared within the configured specific lookback period; calculate a gap between an actual clearing date and a pattern date predicted by the date-shift-payment pattern feature for the filtered closed invoices;

computing a revenue concentration level within the determined adherence evaluation window by calculating a cumulative sum of the calculated gaps based on revenue percentage; and

selecting the optimal pattern with a maximum revenue concentration level in the adherence evaluation window for each granularity level instance;

predicting a financial transaction payment pattern for the set of open invoices based on the optimal pattern, the set of payment pattern features and the set of granularity level instances by using a financial-transaction-pattern-prediction-based ML model, wherein the financial transaction payment pattern comprises an estimated date at which the one or more customers and the one or more vendors complete a financial transaction to clear the set of open invoices wherein the financial-transaction-pattern-prediction based ML model is a regression-based ML model;

pre-processing, using the regression-based ML model, input data comprising the optimal pattern, the set of payment pattern features, and the set of granularity level instances to ensure a format suitable for input into the regression-based ML model;

training a set of decision trees based on the pre-processed input data to iteratively correct prediction errors for the financial transaction payment pattern, using the regression-based ML model;

applying at least one of: regularization to prevent overfitting, and encoding of categorical features to eliminate need for extensive pre-processing, using the regression-based ML model;

outputting a weighted average of the predictions from all the set of decision trees, using the regression-based ML model;

dynamically monitor prediction performance of the regression-based ML model in real time;

identifying instances where the prediction performance decreases below a predetermined threshold;

automatically generating new training data based on the identified instances;

retrain the regression-based ML model using newly generated training data; and

outputting the predicted financial transaction payment pattern on the user interface screen of one or more electronic devices associated with an user.