The Role of Intelligent Agents in DSS Data Mining Applications - Potential Benefits and Pitfall - Case Study Example

The Role of Intelligent Agents in DSS Data Mining Applications: Potential Benefits & Pitfalls Introduction Foster, McGregor, and El-Masri (2005: 2) define an intelligent agent as “a system that perceives its environment and acts upon the information it perceives.” Kirkos and Manolopoulos (2004:1) describe data mining, which can also be called Knowledge Discovery in Databases, as having the objective to discover valid, complex, and hidden information in large bodies of data. These concepts are combined in this paper with Decision Support Systems, which are systems designed to gather and organise data to assist in decision-making. The paper is organised as follows: First, a brief description of different kinds of decision support systems is given. A description of data mining processes follows, and then an explanation of what intelligent agents are and how they work in data mining is given. Finally, some examples of decision supports systems that use intelligent-agent data mining processes are examined, with a particular focus on the benefits and shortcomings of each. Decision Support Systems Many different kinds of Decision Support Systems (DSS) exist, but in order to be effective every DSS must have certain features. Most importantly, the DSS must be specific to the application in which it is used; a DSS that works for a manufacturing operation, for example, is not the same as one which would be used for marketing. A DSS must also be able to process different kinds of data, different groups of data as the user requires, present it in a variety of ways, and offer more than one alternative in how to implement the results. (Holsapple & Whinston, 1996: 144-145) The DSS requires human interaction; as its name suggests it supports decision-making, and is not substituting for the human responsibility for making decisions. Nevertheless, the assistance of DSS is valuable in many different kinds of businesses. In the pharmaceutical industry, enormous amounts of data from clinical testing of different medicines or chemical compounds needs to be sorted out, sometimes over a long period of years, in order to efficiently plan further testing and manufacture. (Ranjan, 2009) In the medical field, a form of DSS known as an Intelligent Decision Support System (IDSS) helps doctors and nurses plan treatment for patients. Rather than functioning as most DSS by presenting one or more certain alternative courses of action, however, the IDSS presents treatment options in terms of probabilities of success. (Foster, McGregor, & El-Masri, 2005: 4) In fields such as accounting, finance, and e-commerce, DSS is useful for different reasons. Transactions in financial markets occur quickly, often in real-time across different time zones. By the time a market analyst or trader can gather and assess the data he needs to make a judgment, the opportunity to use it may well have already passed. Financial transactions in the fast-paced world of modern markets are also susceptible to fraud and errors; this is one of the most useful areas in which intelligent agents and data mining can be applied. (Mraovic, 2008) DSS can also be applied in a customer-oriented rather than a business-oriented role. In e-commerce applications, where direct interaction with customers is limited, DSS can be applied as a customer decision-supporting tool, gathering and organising information from production, warehousing, and distribution inputs to present customers with a number of purchase options and conditions. (Lee & Park, 2003) How intelligent agents in data mining can be used with DSS in these kinds of scenarios will be discussed in following sections. Data Mining & How It Works Data Mining (DM) in its most simple definition is the discovery of relevant data from large sets of information. (Ranjan, 2009) ‘Relevant data’ can take two forms: tacit knowledge, which can otherwise be described as ‘experience’ and is the knowledge and skills people develop over time spent in a particular effort; and explicit knowledge, which is factual and can be transferred easily. (Wild & Griggs, 2008: 5) The task which users are trying to accomplish is ‘knowledge synthesis’, the processing of ‘information’ into ‘knowledge’ which can be used to make a decision; data mining helps this process by discovering patterns in bodies of information. (Mohamed, Stankosky, & Mohamed, 2009) The potential problem is that the amount of tacit knowledge held by people differs from one person to another. People know different types of information, and know different amounts of information based on the length of their experience and their mental abilities. Thus, two people presented with the same level of explicit knowledge will draw different conclusions, because they are synthesising that with different sorts of tacit knowledge. Data mining helps by discovering patterns of tacit knowledge within the bodies of information people use, reducing some of the differences between what different people have in their own minds. It does this using a number of techniques, such as (Kirkos & Manolopoulos, 2004: 2-4): Neural Networks – A form of data management that works similar to the way the human brain works, with mathematical rules governing the ‘firing of neurons’ that move bits of data. A kind of ‘black box’ technology that can be confusing, the way in which the patterns developed in the system can be understood is through algorithms. (Kirkos & Manolopoulos, 2004: 3) Genetic Algorithms – Algorithms that function by forming rules as strings of data; the strings with the highest fitness, i.e., greatest applicability and most likely to be true, are then replicated and spread. (Kirkos & Manolopoulos, 2004: 3-4) This helps the system find specific patterns in the data. A simple example would be: if A and B exists, then C is true. If this rule has high fitness, in other words, if the probability of A and B existing is high, then the rule would be duplicated, and every instance of A and B = C in the data would be found. Decision Trees – Decision trees are a means of finding differences between observed pieces of data. The data is broken down into smaller and smaller sets until no difference exists between the subsets. The degree of the separation – how many steps and how many subsets remain at the end – helps to establish patterns that can be found in other parts of the data collection. Rough Set Theory – This is a way of determining associations and dependencies of different pieces of data. A “rough set” might consist of, for example, all items definitely belonging to a certain Set A and all items that cannot be defined as not belonging to Set A. Further association and attribution rules can be applied to the rough set to narrow its connections even further. (Kirkos & Manolopoulos, 2004: 4) Case Base Reasoning – This is a method that most closely matches tacit knowledge, and is probably easiest for most people to understand. The system searches for similar cases in the data to compare to the present case; those that match most closely are the ones presented as the possible solutions. One thing that is obvious from these few examples is that DM processes and human thought processes are similar. In DM applied to DSS, the objective is to use these intuitive features to find the needed information in a large amount of data, especially the information that the human user does not realise he needs. Intelligent agents in DM can refine this process even further. Intelligent Agents & How They Work An agent is a system or entity that has the ability to act on behalf of another, either by a set of directions or autonomously. An intelligent agent (IA) in data mining makes use of that autonomous option: it is a system which can perceive its environment, and act upon the information it perceives. (Foster, McGregor, & El-Masri, 2005: 2) It is helpful to understand how an IA works by thinking of a decision solution as a task, one which has a certain degree of complexity. Wang and Wang (2005: 3) describe four key characteristics of a business process or task: 1. The task requires a number of decisions: The ultimate, final decision representing the problem to be solved requires that a series of lesser decisions be made on the way to arriving at the final solution. 2. The decisions are interdependent: The decisions may occur sequentially or simultaneously, but each decision has some effect on all the other decisions. 3. The environment of the decisions changes as a result of the decisions, and can also change on its own: The conditions that exist at the beginning of the problem change as the subordinate decisions are made, so that the final decision is not based on the original set of conditions, but on the last set of conditions. 4. The decisions are made in real-time: The decisions all work in a ‘forward’ direction; if a decision is incorrect, it has already changed the environment so that it is impossible to back-track to the exact conditions before the incorrect decision and try a different option that was rejected at that time. An IA is able to perceive the changing conditions as pieces of information are included or discarded in the overall solution, which it does according to one of the processes described in the previous section (neural networks, genetic algorithms, decision trees, rough sets, or case base theory). In other words, the DM which employs an IA is able to change its selection of information based on the changes in specifications that are caused by the selection of earlier pieces of information. The Pros & Cons of IA Applications in Data Mining-Supported DSS Data mining is an improvement over systems used for less-refined DSS (such as OLAP), in that it is not limited by the need for a specific and exact question or hypothesis to be asked. (Hedelin & Allwood, 2002) DM instead can search for trends or patterns, which can suggest the hypothesis or question after the fact, in a manner of speaking. In this way, DSS using DM is a more flexible and thorough system than a DSS which does not use DM. In general, a system using DM requires less human intervention and analysis, depending on how specific and detailed the problem put to the system is at the beginning; less detail, of course, would require greater analysis. DM employed in this manner, however, assumes a static set of conditions; patterns and trends do not change, nor does the underlying environment change as a result of decision selections. When IA is added, the dynamic nature of the data environment is taken into account. Decisions – such as those in the medical field, for example – where a degree of uncertainty about the outcome cannot be avoided can be offered more consistently as a set of alternatives with varying levels of confidence. This provides a richer and more detailed set of information for experts (such as doctors) to assess and make a choice. (Foster, McGregor, & El-Masri, 2005: 4) These decisions then become the basis for a better set of organisational knowledge, which improves the decision process in subsequent problems. (Zhang, 2009) IA-driven DM in decision support systems is not a perfect solution in all cases, however. For environments that involve many kinds of different work tasks and dynamic, fast-operating information paths, IA is a good solution because it works to identify patterns of logic in complex business activities. (Wang & Wang, 2005: 9) Good examples of the kinds of enterprises where IA can be helpful are the medical field, financial businesses, and complex manufacturing operations that rely on long, interconnected supply chains. In less-complex environments where there are relative few work tasks and the movement of information is reasonably straightforward, however, an IA system may be unnecessarily complex and actually slow down work processes. The other significant negative characteristic about IA in relation to its use by business enterprises is, ironically, its sophistication. Studies have shown that the acceptance of business process and information technology is inversely proportional to its complexity and the ease with which it can be understood by users. (Hedelin & Allwood, 2002, and Kim & Trimi, 2007) Because of the thoroughness with which they can find and retrieve information, DM systems, and particular IA-driven DM systems, can be misused to gather private and secured information. (Mraovic, 2008) In addition, IA-enhanced DSS systems require a high degree of human expertise and intervention, particularly in applications where recommended decisions are expressed in terms of probabilities or degrees of uncertainty, such as in the medical field. The problems these factors can cause are obvious. If managers do not understand the capabilities of the system and lack a clear understanding of how it operates, they will have less confidence in using it. If the manager is not confident and committed to the system, he will not have much success in promoting its adoption and use throughout the rest of the organisation. The confidence factor applies to the security concerns as well; users will either be hesitant to use the system to its full capabilities, or attempt to impose security safeguards that may interfere with its proper function. And finally, if the users do not understand that the system is an additional tool to use in their decision-making process, rather than a substitute for it, there is a strong likelihood it will render unsatisfactory results. Conclusion Decision support systems serve as an additional tool that managers can use in planning, problem-solving, and decision-making. But most basic DSS systems require a specific problem or question as an input before they can develop recommendations. Data mining can be applied to DSS to find patterns in large sets of information, which is helpful when a problem is not clearly understood or developed; in other cases, DM is effective in finding information and data connections that might not be apparent to humans alone or lesser systems. Basic DM has one drawback, however, in that it does not necessarily work well in dynamic environments where conditions change. In most real-world situations, one small decision changes the environment of the entire problem, and consequently changes the nature of the data needed to make subsequent decisions. Intelligent agents in DM solve this problem by being able to perceive their environment, and alter their data selections in real-time in response to changes caused by other actions. IA provides a more detailed and accurate solution, and is particularly useful in circumstances where outcomes may be uncertain, such as in medical treatment. There are, however, a few drawbacks to IA-driven systems. Skilled human interaction is still required, not only to interpret and verify the results, but in understanding and using the system properly. IA might not be suitable for all businesses; in less-complex environments, it may actually increase the time needed for the decision process. And there are potential privacy and security concerns if the system is used improperly. But with proper understanding, application, and confidence in the system, the advantages of IA outweigh its disadvantages in most circumstances, and can yield timely and effective results. Works Cited Foster, D., McGregor, C., and El-Masri, S. (2005) “A Survey of Agent-Based Intelligent Decision Support Systems to Support Clinical Management and Research”. [Internet] University of Western Sydney, June 2005. http://www.diee.unica.it/biomed05/pdf/W22-102.pdf. Hedelin, L., and Allwood, C.M. (2002) “IT and strategic decision making”. Industrial Management & Data Systems, 102(3): 125-139. [Internet] Emerald: http://www.emeraldinsight.com/10.1108/02635570210421318. Holsapple, C.W., and Whinston, A.B. (1996) Decision Support Systems: A Knowledge Based Approach, 10th ed. Los Angeles: West Group. Kim, S., and Trimi, S. (2007) “IT for KM in the management consulting industry”. Journal of Knowledge Management, 11(3): 145-155. [Internet] Emerald: http://www.emeraldinsight.com/10.1108/13673270710752162. Kirkos, E., and Manolopoulis, Y. (2004) “Data Mining in Finance and Accounting: A Review of Current Research Trends”. [Internet] Proceedings of the 1st International Conference on Enterprise Systems and Accounting, Thessaloniki, Greece. http://delab.csd.auth.gr/papers/ICESA04km.pdf. Lee, J.H., and Park, S.C. (2003) “Agent and data mining based decision support system and its adaptation to a new customer-centric electronic commerce”. Expert Systems with Applications, 25(4): 619-635. [Internet] ScienceDirect: doi:10.1016/S0957-4174(03)00101-5. Mohamed, M., Stankosky, M., and Mohamed, M. (2009) “An empirical assessment of knowledge management criticality for sustainable development”. Journal of Knowledge Management, 13(5): 271-286. [Internet] Emerald: www.emeraldinsight.com/10.1108/13673270910988105. Mraovic, B. (2008) “Relevance of Data Mining for Accounting: Social Implications”. Social Responsibility Journal, 4(4): 439-455. [Internet] Emerald: http://www.emeraldinsight.com/10.1108/17471110810909858. Ranjan, J. (2009) “Data mining in pharma sector: benefits”. International Journal of Health Care Quality Assurance, 22(1): 82-92. [Internet] Emerald: http://www.emeraldinsight.com/10.1108/09526860910927970. Wang, M., and Wang, H. (2005) “Intelligent Agent Supported Business Process Management”. Proceedings of 38th Hawaii International Conference on System Science. [Internet] CiteSeer: http://citeseerx.ist.psu.edu/viewdoc/summary?doi=10.1.1.112.2470. Wild, R., and Griggs, K. (2008) “A model of information technology opportunities for facilitating the practice of knowledge management”. VINE: The journal of information and knowledge management systems, 38(4): 490-506. [Internet] Emerald: http://www.emeraldinsight.com/10.1108/03055720810917732. Zhang, Z. (2009) “Personalising Organisational Knowledge and Organisationalising Personal Knowledge”. Online Information Review, 33(2): 237-256. [Internet] Emerald: http://www.emeraldinsight.com/10.1108/14684520910951195. Read More