[edit] Brief Description

[edit] Wiki description

General

What does predictive analytics mean?

Branch of business intelligence category, predictive analytics uses data mining and statistics to make predictions on future happenings. The predictions tell you what are the odds that a certain event will be taking place or not, under what circumstances, or following trends

Usually, software tools involve rigorous data analysis to find patterns in historical and transactional information that can help them identify risks or opportunities. They find relationships in the factors that are assessed to guide decision made by a company management by giving a certain set of condition a "predictor". The predictor is a variable that can be assigned to a individual or any other entity to identify its risk or potential and predict its future behavior. In financial systems, credit scoring is used to asses a persons probability of paying in time future credit. The predictors in this case are the individual credit history, income, family members, loan applications and so on. Another popular field where predictive analytics are used is insurance companies. They take into consideration age, gender, driving record before issuing a policy

Current use of predictive analysis

The variety of applications where prediction analysis can be applied is endless. In the enterprise world, the success rate of prediction analytical tools is mostly notable in the CRM area but there are also some other fields that are worth mentioned

Customer Relationship Management

If you wonder if your marketing campaigns can be improved with a software tool you may find out that prediction analytics can do much more. Many predictive analytics techniques are applied to customer data to pursue CRM objectives, CRM being the most frequent commercial application of this type of business intelligence tools

The central piece, the predictor, in the CRM can be the recency (time since the customer last purchased something), customers time online on your website, monthly usage and so on. The purpose of predictive analytics is to combine this predictors for smarter rankings. The best combination is not easy to find, and that's where predictive analytical tools come in handy to develop the predictive model from the customer data that increases your revenues the most

The predictive analytics tools build the optimal model automatically, but the overall process to mange and integrate this software is not automatic - you do need the marketing expertise. Customer Relationship Management software tools (both free and commercial) with user reviews are covered on ITerating and are classified under the subclasses :

• Customer Service
• Call Center
• Help desk
• Sales
• Contact Center
• Marketing

Here are some examples where predictive analytics give you the best results in the customer managing process:

• Direct Marketing - Because competition is high you need to increase the response rate of your customers. Predictive analytics understands the amount of variability and tailors the marketing strategy for better results. For that it take into consideration customer data like past purchasing history, past response rates, demographic etc
• Customer retention To maximize consumer satisfaction in a very competitive market, the likelihood of a customer loss can be predicted and prevented
• Cross-sell For efficient cross selling predictors like spending, usage and history are analyzed
• New markets Using business intelligence modeling, new business models can be design to enter new markets
• Price estimation Agent-based models can find the right parameters for business models (membership fees, , annual fees, emergency fees)

Human resources

Tracking employees performance is perhaps the most critical part of Business Performance Management. Predictive analytics provide the tools to better identify the sources of inefficiency and plan corrective actions. Predictive Analytics can also be used for HR planning, helping managers anticipate which skills will be required in the future and in what quantity, with implications on hiring strategy as well as internal mobility.

Warehouse management

Portfolio, product or economy level prediction can be made with analytics methods. Distribution of products can be optimized by predicting store level demand for better management of inventories . Adoption of new products can be also accelerated using agent-based models on existing research data

Financial services

Scoring models process a customer’s credit history, loan application, customer data, etc., in order to rank-order individuals by their likelihood of making future credit payments on time

• Underwriting A financial company needs to assess a borrower’s potential and ability to pay before granting a loan. Predictive analytics can underwrite these quantities to better predict the chances of on time payment, bankruptcy and so on

• New services The reaction and the likelihood of a customer adoption to a new fee/new service can be predicted. Proper predictive analytics can answer questions like: will the customers use the service, will they refuse to pay for it, will they stop using all your services?

• Collection analytics The cost of collection process can be reduced with the right resource allocation

Insurance

Insurance carriers face many challenges in the areas of operational risk, market risk, customer profitability, fraud and ratemaking

Predictive analytic tools can empower the insurance industry’s marketing, sales and risk management departments to create revenue opportunities, and reduce risks of loss, at every stage of the customer credit lifecycle, through cost effectively integrate predictive capabilities with their underwriting and pricing engines, to provide comprehensive analysis and surveillance capabilities of their claims systems for retrospective and prospective fraud and abuse detection

Fraud is an important loss source within online business especially but False insurance claims, fraudulent transactions, inaccurate applications can be reduced with the means of predictive tools

Telecommunications

In telecommunications, predictive analytical models can design network creation and routing policies for distributed data storage. They can simulate network traffic, analyze system under many scenarios and predict its behavior

Travel

Efficiency in airport can be predicted with just one variable change, also flight tickets prices can be foreseen. If you want to know if you should wait another day before you fly somewhere as the prices can be much lower, check Farecast out -an online services that uses predictive analytics to show you best prices available in the future

Scientists focus on predictive analytics for space travel as data mining and probabilistic risk assessment will help improve safety. For instance one engineer using a BI tools can now identify problems in the aircraft structure in three to four hours, where it once took eight engineers six weeks

Healthcare

In the healthcare industry, predictive analysis can be applied in many ways. One of the most popular is the health insurance enrollment, where customers are overwhelmed by number of choices. In such a high competitive environment, models can be applied to increase prediction accuracy for health plan registering. Also, for a health insurance provider, predictive analytics can analyze a few years of past medical claims data, as well as lab, pharmacy and other records where available, to predict how expensive an enrollee is likely to be in the future

Pharmaceuticals

Pharmaceutical companies are the largest customers of predictive analytics and data mining in general and there are many ways in which they use the BI tools:

• Drugs Adoption -TheNPV of early-stage compound can be evaluated under wide range of possible scenarios: competitive, regulatory, marketing, market adoption. Traditional aggregate-level, static, linear models and assumptions built to forecast an environment that is fundamentally dynamic and non-linear

• Test new therapeutic hypotheses- Prediction of proteins behavior and traffic to test new healing possibilities for different diseases

Statistical techniques used by predictive software tools

Basically, predictive modeling is the process of creating or selecting a model to predict the likelihood of an event or outcome A model, a mathematical one is an abstract model that uses mathematical language to describe the behavior of a system. Mathematical models typically focus on a specific functional form or process (linear models, generalized linear models, hidden Markov models)

To use a mathematical model, you have to follow some steps:

• Determine what model type to use
• Determine what data to use
• Adjust model parameters to fit the available data
• Extrapolate to estimate future outcomes

Regression models

Regression analysis represents the basic technique used in predictive modeling. To determine the mathematical equation between different variables, different approaches are used

Linear regression model -using the ordinary least squares estimation Discrete choice models Logistic regression Multinomial logistic regression Probit regression

Time series models

Time series techniques are used to predict future behavior of variables as regression models can not be used in forecasting predictable components in the series Time series models compute difference equations with stochastic factors. Two commonly used forms of these models are :

• autoregressive models (AR)
• moving average (MA) models
• autoregressive moving average (ARMA)

Latest time models:

• ARCH (autoregressive conditional heteroskedasticity)
• GARCH (generalized autoregressive conditional heteroskedasticity)

Survival or duration analysis- does the analysis of events, calculating their chance of survival. It uses hazard rate to measure the probability that the event will occur at time t conditional on surviving until time t

Classification and regression trees methods are used to categorize the data. Most popular method is Leo Breiman's Random forests

Machine learning

Machine learning was developed to enable computers to learn.Today there are many methods of advance statistical machine learning for learning and classification. Basically, it emulates human cognition and learn from training examples to predict future events

Most common methods used:

• Neural networks
• Radial basis functions
• Support vector machines
• Naive Bayes
• k-nearest neighbours

Predictive Analytics software tools

Popular tools used

Commercial:

• Insightful Miner
• Enterprise Miner by SAS
• TM1 solution from Cognos
• BusinessObjects XI

Open source:

• Weka
• RapidMiner

How to choose the right software for your business

Events

• Prediction Impact 2008 -Predictive Analytics for Business, Marketing and Web, Training Seminar. Locations : Toronto (April), San Francisco (May), New York City (June)
• Business Forecasting Conference Embassy Suites Hotel Cary NC June 4 2008

See also

• Learn more about Business Intelligence Software
• Business Intelligence Tools with Users Reviews
• Agent based modeling software - free and commercial