Real-time Transaction Compliance Audit and Monitoring

As more and more transactions are processed with less and less human intervention, it is essential to have intelligent compliance audit and monitoring systems in place. Compliance audit and monitoring of business transactions can occur for various reasons;

• Regulatory compliance: laws and government regulations require certain controls imposed on business transactions, e.g., SarBox Sarbanes-Oxley Act Section 404, etc.
• Internal policies: Companies and organizations set out internal policies for management and marketing purposes.
• Risk management: Credit payments have risk of being defaulted. Detecting early signs of default may prevent further credits.
• Business Activity Monitoring (BAM): Monitor business activities and detect early signs of risk and opportunities.
• Verification: Transactions can be mistakenly processed. Verification can reduce mistakes and errors.
• Fraud detection: Fraudulent claims are costly for healthcare and insurance industries. Early detection can be vital in reducing fraudulent claims. For more, please read Healthcare Fraud Detection.
• Event monitoring: Detection of unusual events is important in various industries.
• Customer churn alert: detection of imminent customer churns.
• Intelligent precision marketing: Intelligent algorithms can detect cross-sell and up-sell opportunities automatically as customers select products.

Predictive Modeling by Machine Learning

Predictive modeling by machine learning techniques can be used to audit and detect risky transactions. For example, neural network can be trained so that it can detect potentially fraudulent credit purchases, or to alert risky insurance applications to actuaries, and so on.

Predictive models are developed from past historical data using machine learning techniques. CMSR Data Miner supports the following predictive modeling tools;

• Neural Network and Deep Learning
  Neural network is a predictive model which is based on the architecture of, say, our brains. Neural networks can be trained to predict both numerical risk levels and categorical risk classifications. It can predict rare events very well. In addition, multiple models can be used for deep learning. For more, please read Neural Network.

• Decision Tree and Risk Classification
  Decision tree is a predictive modeling technique which is based on recursive segmentation and probabilistic reasoning. It is primarily used to classify risk categories. CMSR Data Miner supports prediction of statistical probability as well. For more, please read, Decision Tree Classifier.

• Regression and Time-series
  Regression develops mathematical predictive formulas for numerical information. Although limited for general modeling, it can be used in deep learning as component models.

For risk management, please read;

• Credit Risk Scoring and Predictive Modeling
• Insurance Predictive Modeling and Risk Scoring

Machine Intelligence and Audit/Monitoring Rule Engine

RME-EP is a new generation expert system shell rule engine that combines the best of Rule-based Inference and Predictive Modeling. RME-EP has powerful rule specification language which is based on SQL database query language. Complex rules can be specified very easily. In addition, RME-EP can be used to implement Deep Learning models.

For more information, please read Rule Engine with Machine Learning, Deep Learning, Neural Network.

For software and download, please read CMSR Data Miner.

RME-EP Audit Rule Specification Language

RME-EP provides a very powerful audit/monitoring rule specification language. It is based on SQL-99 and very easy to learn and intuitive. The followings show example codings of rules. They are sales trend KPI scan and monitoring rules.

/* 
 * To find series that fluctuate over the specified, say, less than 0.1, coefficient 
 * of determination or over 90% of coefficient of variance.
 */
// /*RR=*/ Maximum coefficient of determination, say, 0.3
// /*CV=*/ Minimum coefficient of variance of linear regression, say, over 50%
DECLARE "State-Fluctuation"  AS STRING OUTPUT INITIAL VALUE 'No'; 
RULE FluctuationDetection:
IF TIMESERIES(RR, (LINEAR, EXPONENTIAL), 1, "Series", "Current") < 0.3
	OR TIMESERIES(CV, (LINEAR), 1, "Series", "Current") > 50.0 THEN
	SET "State-Fluctuation" AS 'Yes'
END;

/*
 * If the difference between the current value and the predicted value is 
 * bigger than rate-times of the regression average absolute errors.
 */
// /*TR=*/ times rate of average absolute errors, say, 1.5
DECLARE "State-BigJump"  AS STRING OUTPUT INITIAL VALUE 'No';
RULE BigJumpDetection:
IF ABS("Current" - TIMESERIES(1, (LINEAR, EXPONENTIAL), 1, "Series")) > 
	TIMESERIES(AVG_ABS_ERROR, (LINEAR, EXPONENTIAL), 1, "Series") 
		*  1.5 THEN
	SET "State-BigJump" AS 'Yes'
END;

/*
 * Solid linear growth over 3% in average with RR > 0.7.
 */
// /*RR=*/ Miniimum coefficient of determination, say, 0.7
// /*LR=*/  Minimum linear growth rate 3% = 0.03
DECLARE "State-SlodLinearGrowth"  AS STRING OUTPUT INITIAL VALUE 'No'; 
RULE SolidLinearGrowth:
IF TIMESERIES(RR,  LINEAR, 1, "Series", "Current") >  /*RR=*/ 0.7 AND
	TIMESERIES(GROWTH_RATE,  LINEAR, 1, "Series", "Current") 
		>  0.03 THEN
	SET "State-SlodLinearGrowth" AS 'Yes'
END;

/*
 * Solid linear decline over 3% in average with RR > 0.7.
 */
// /*RR=*/  Miniimum coefficient of determination, say, 0.7
// /*LR=*/  Maximum linear decline rate -3% = -0.03
DECLARE "State-SlodLinearDecline"  AS STRING OUTPUT INITIAL VALUE 'No'; 
RULE SolidLinearDecline:
IF TIMESERIES(RR,  LINEAR, 1, "Series", "Current") >  /*RR=*/ 0.7 AND
	TIMESERIES(GROWTH_RATE,  LINEAR, 1, "Series", "Current") 
		< -0.03  THEN
	SET "State-SlodLinearDecline" AS 'Yes'
END;

/*
 * Solid exponential growth with RR > 0.8.
 */
// /*RR=*/ Minimum coefficient of determination of exponential regression
// /*ER=*/ Minimum exponential growth rate, say 1.01
DECLARE "State-SlodExponetialGrowth"  
		AS STRING OUTPUT INITIAL VALUE 'No';
RULE SolidExponentialGrowth:
IF TIMESERIES(RR,  EXPONENTIAL, 1, "Series", "Current") >  /*RR=*/ 0.8 AND
	EXP(TIMESERIES(GROWTH_GRADIENT, EXPONENTIAL, 1, 
		"Series", "Current"))  > 1.01 THEN
	SET "State-SlodExponetialGrowth" AS 'Yes'
END;

Business Activity Monitoring (BAM)

Business Activity Monitoring (BAM) has been for a while. However, it is limited to real-time dashboard applications. Basically, it displays visual real-time graphs and charts based on KPIs or Key Performance Indicators. It lacks intelligent predictive power that businesses may require. RME-EP rule engine with predictive models can change this. KPIs can be used as predictors. Predictive models can be developed for KPIs. RME-EP models based on deep learning techniques can be implemented.

Rosella BI Server for Web Deployment

Rosella BI Server is a Java J2EE based web service platform. RME-EP models which include predictive models can be deployed on BI server. In addition, business applications can incorporate RME-EP audit models using HTTP/JSON requests or using JSP pages. BI Server is a very efficient high performance system that can process over 5,000 HTTP/JSON requests per second.

For more information, please read Rosella BI Server.