Week 8, Lecture 17 and Lecture 18 Predictive analytics [Page 105] Predictive analytics is a highly computational data-mining technology that uses information and business intelligence to build a predictive model for a given business application. So, predictive analytics is all about using historical information to predict future events and outcomes. Predictive analytics has application in a wide variety of industries including insurance, retail, healthcare, travel, and financial services. Within each of those, the business applications will obviously vary but include applications like customer relationship management, supply chain management, and credit scoring (a very popular predictive analytics application within financial services for determining the likelihood a credit applicant will be able to make future payments). Predictive analytics includes a prediction goal and many prediction indicators. A prediction goal is the question you want to address by the predictive analytics model. Predictive goals might include: o Which suppliers are most likely to deliver raw materials in the next six months with a defective rate higher than 0.001%? o What customers are mostly likely to respond to a social-media campaign within 30 days by purchasing at least two products in the advertised product line? Defining the right prediction goal is key. If you build a predictive analytics model for a prediction goal that is in some way wrong, you can expect your subsequent activities to be erroneous (i.e., a failure) as well. A predictive indicator is a specific measureable value based on an attribute of the entity under consideration. The best way to understand prediction goals and prediction indicators is to give example about them. Let us focus on the second prediction goal listed above (i.e. what customers are mostly likely ). This is a CRM application which will use customer profile information, customer history information, and campaign history information (see Figure 4.6). Dr. Anas Aloudat Page 1 of 7
The predictive analytics engine will use many pieces of information from which it will develop a set of prediction indicators that produce the best model for predicting the customer behavior you seek. Figure 4.6: Analytics Process of Customer Prediction In the end, the predictive analytics engine produces a predictive model as a formula for each customer that provides a single predictive value by which customers can be ranked from most to least likely to respond to a social-media campaign within 30 days by purchasing at least two products in the advertised product line. As we said earlier, a prediction indicator is a measureable value based on an attribute of the entity under consideration. Regarding the entity customer, the predictive analytics engine may determine that the following attributes are important for making the right prediction: o Frequency of purchases o Proximity of data of last purchase (called recency) o Presence of Facebook (Yes = 1; No = 0) o Presence of Twitter (Yes = 1; No = 0) o Number of purchases with multiple products Then, the predictive analytics engine will assign weights to each prediction indicator that represent the relative significance of each indicator a compared to the other indicators. This is the highly computational aspect of predictive analytics and something certainly beyond the scope of this book. Dr. Anas Aloudat Page 2 of 7
Computational models and algorithms used here by the predictive analytics engine include things like linear programming, regression, association rules, market basket analysis, cluster analysis, etc. The predictive model final formula might look like this: Customer# = (Frequency of Purchases in the last 30 days)*.2 + (Recency)*.1 + (Presence on Facebook)*.3 + (Presence on Twitter)*.1 + (Number of Multiple Product Purchases)*.3 As you can see, the predictive analytics engine determined that the indicators of Presence on Facebook and Number of Multiple Purchases were most important, with each receiving a weight of 0.3 (Note that all of the weights sum = 1). Once the formula is applied to all customers, each customer receives an overall predictive indicator score, allowing you to rank order them. Then, it is up to you to determine where to draw the line (i.e. the cutoff point in the list) and identify the subset of customers on whom you will focus your social-media campaign. Text Analytics [Page 108] Text analytics is the process of using statistical, artificial intelligence, and linguistics techniques to convert information content in textual sources like surveys, emails, blogs, and social-media into structured information. Text analytics most definitely falls within the categories of analytics in general and more specifically decision support systems because it works primarily with nonstructured elements, that is, natural languages. Let us consider an example: Gaylord Hotels is using text analytics to better understand the thousands of customer satisfaction surveys it receives every day. The surveys are digitized and fed into a text analytics engine. The engine can identify negative and positive comments, which are then quickly correlated to structured information such as the specific hotel, facilities, amenity services, the room, and employee shifts. With that correlation, employees of Gaylord Hotels can follow up with customers by phone or letter to acknowledge the problem and to offer and apology and even discounts for a return stay. Text analytics is even more complicated and technical than predictive analytics. While predictive analytics relies mainly on statistical models to build predictive models, text analytics makes use of statistical models and also linguistic models. Some of these include: o Lexical analysis the study of word frequency distributions. o Named entity recognition the identification of names for people, places, and things. Dr. Anas Aloudat Page 3 of 7
o Disambiguation the process of determining the specific meaning of a named entity recognition. For example, Ford may refer to a previous U.S. president, and Ford Motor Company, or perhaps a notable person like Harrison Ford. o Coreference the handling of differing noun phrases that refer to the same object. o Sentiment analysis discerning subjective business intelligence such as mood, opinion, and emotion. All of the above fall within the category of linguistic processing of what is often referred to as natural language processing. Professionals in the field of natural language processing have expertise in both linguistics and computer science. Endless Analytics [Page 109] Beyond specialized analytics like predictive analytics and text analytics, you will come across many other analytics applications focused on specific areas of business. While these may use sophisticated data-mining and statistical tools, they are considered broadly within the field of analytics because they focus on fact-based decision making. Analytics here include: o Web analytics the analysis of data related to the Internet, often focusing on optimizing Web page usage. An important subset of Web analytics is search engine optimization (SEO), improving the visibility of a Web site through the use of tags and key terms found by search engines. o HR analytics the analysis of human resource or talent management data for such purposes as work-force capacity planning, training and development, and performance appraisal. o Marketing analytics the analysis of marketing related data to improve the efficiency and effectiveness of marketing efforts including product placement, marketing mix, and customer identification and classification. o CRM analytics the analysis of CRM data to improve functions such as sales force automation and customer service and support. o Social media analytics the analysis of data related to social media use, mainly by customers or competitors, to help an organization better understand the interaction dynamics of itself with its customers and also to help an organization scan social media for competitive intelligence. o Mobile analytics the analysis of data related to the use of mobile devices by customers and employees. Mobile computing and mobile e-commerce are exploding. We will read more about it in Chapter 5. Dr. Anas Aloudat Page 4 of 7
Artificial intelligence [Page 110] Artificial Intelligence (AI) is the science of making machines imitating human thinking and behavior. o Financial analysts use AI systems to manage assets, invest in the stock market, etc. o Hospitals use AI to schedule staff, assign beds to patients, diagnose and treat illness. o Many government agencies including IRS and the armed forces use AI. o Credit card companies use AI to detect credit card fraud. o Insurance companies use AI to spot fraudulent claims. AI lends itself into diverse areas, such as ticket pricing, food preparation, oil exploration, child protection, meteorology, engineering, and aerospace industries. AI is an important tool in the field of analytics. AI can be independent, stand-alone decision making systems, or they can be embedded into larger analytics systems, carrying out and executing specific functions. AI are usually classified into one of the following four main categories: 1. Expert systems 2. Neural networks and fuzzy logic 3. Genetic algorithms 4. Agent-based technologies Expert Systems [Page110] Expert system (also called knowledge-based system) is an AI system that applies reasoning capabilities to reach a conclusion. Expert systems are excellent for diagnostic and prescriptive problems. Diagnostic problems (تشخيصية) are those requiring an answer to the question: What is wrong? and correspond to the intelligence phase (Phase 1) of decision making. Prescriptive problems (وصف) are those that require an answer to the question: What to do?, and correspond to the choice phase of decision making..(نطاق) An expert system is usually built for a specific applications area called a domain Examples of domains and their expert systems: o Accounting: for auditing, tax planning, management consulting, and training. Dr. Anas Aloudat Page 5 of 7
o Medicine: to prescribe antibiotics where many considerations must be taken into account first (such as the patient s medical history, age, the source of infection, the price of available drugs). o Human resource management: to help personnel managers determine whether they are in compliance with national employment laws. o Forestry management to help with harvesting timber on forest lands. A simple example of an expert system is to tell a driver what to do when approaching a traffic light. See Figure 4.6. In the example above, there is a recurring problem and to solve it you need a well-defined set of steps. An expert system is well-suited for such kinds of problems. You ve probably gone through the question-and-answer set hundreds of times if you are a driver without even realizing it. Expert systems use rules, very similar to the association of rule or dependency modeling concept we discussed earlier. An expert system can be a stand-alone application like the one a doctor would use to aid in the diagnosis of a problem, or it may be embedded into an analytics system. Dr. Anas Aloudat Page 6 of 7
Neural Networks and Fuzzy Logic [Page 111] A Neural network (NN) (called also an artificial neural network or ANN) is an AI system that is capable of finding and differentiating patterns: o An ANN can learn by example and can adapt to new concepts and knowledge. NNs are widely used for visual pattern and speech recognition systems: o Example: Smartphones and PDA that decipher your handwriting are probably a NN that analyzed the characters you wrote. NNs are also used in a variety of situations. For examples: o Bomb detection: in U.S. airports NNs are used to sense trace elements in the air that may indicate the presence of explosives. o Chicago Police Department use NNs to detect corruption (فساد) within its ranks. مخططات o In medicine, NNs are used to check 50 million electrocardiograms (ECG per year, checking for drug interactions and detect anomalies in tissue (القلب samples that may signify the onset of cancer and other diseases. NNs can detect heart attacks and even differentiate between the subtly different symptoms of heart attacks in men and women. o In business, NNs are used for securities trading, fraud detection, evaluating loan.(تقييم ( appraisal applications, target marketing, and real estate o NNs are used to adjust temperature settings, control machinery, and detect malfunctioning machinery. NNs are most useful for identification, classification, and prediction when a vast amount of information is available. By examining hundreds or even thousands of examples, a NN detects important relationships and patterns in the information: o For example, if a NN is provided with the details with numerous credit card transactions and was told which ones are fraudulent, eventually it will learn to identify suspicious transaction patterns. Dr. Anas Aloudat Page 7 of 7