Combining modelling and DSS for effective business analytics

PKeenan_post

By Peter Keenan, Associate Professor at Centre for Business Analytics, University College Dublin.

Modelling has always taken advantage of the power of computing, and the enormous increase in computing power has allowed the use of ever more sophisticated modelling techniques. However, despite this growing computer power, modelling approaches have often only been capable of solving simplified versions of problems. While this simplification makes the problems computationally tractable, these simplifications often make theoretically efficient solutions of little value in the real world. To provide effective solutions for Business Analytics, we are interested not only in the theoretical power of the solution technique, but also in its actual fit with the real business problem. Faster computers will simply lead to an inappropriate answer in less time, if the problem modelled is not a useful representation of the one we actually want to solve.

What effective business analytics requires is the application of realistic modelling to appropriate data and the presentation of the output in a form useful to business decision-makers. This wish to provide a system under the control of decision-makers was the original inspiration for Decision Support Systems (DSS), which combined models, databases and user friendly interfaces to support decisions. The idea of DSS was to allow users, business decision-makers who understood the decision, interact with and control the modelling process and understand the solution presented. The greatest benefit of DSS is obtained when the scope of the problem is sufficiently complex for the decision-maker to have problems fully understanding it, and so where modelling can assist with problem solving. DSS can support such problems where the problem and solution can also be effectively visualised, allowing the user effectively intervene in the solution. Such user intervention is particularly valuable where the modelling approach does not include all of the practical constraints of interest to the decision-maker. Transportation applications were an early area of DSS application and problems like vehicle routing are both diverse and complex. However, the user can visualise the problem on a map interface and intervene in the solution, for instance by moving a customer from one route to another. There is a very wide range of decisions and so a very diverse range of DSS applications in all business domains.

DSS first emerged in the 1970s, when databases and interactive interfaces first became available. As technology has evolved, DSS has taken advantage of new developments to improve decision support. These developments have included high resolution graphics, mobile devices, and virtual reality interfaces. New technology means that better decision support can be provided. For instance, in the vehicle routing field we now have comprehensive spatial databases of road networks, geographic information system (GIS) software and global positioning system (GPS) devices to record the position of vehicles. These can be integrated in a modern DSS to enhance the value of long established modelling approaches so that solution quality is improved by more realistic modelling. If analytics modelling is to be an applied science, rather than a purely theoretical exercise, then models must be embedded in systems that actually facilitate real decisions; this is the field of DSS.

Anuncios

Optimization on environmental models: Formal co-simulation to find a sustainable design

PFonseca_post

By Pau Fonseca i Casas, professor, Polytechnical University of Catalonia.

Environmental simulation models are difficult to handle due to the inherent complexity of the elements involved on the analyzed systems. How we can deal with the complexity of those systems? How can we work with the diversity of the languages and terminologies used by different experts? How can we allow an extensive Validation and Verification of such models? The use of formal languages and co-simulation techniques allow the combination of several simulation and optimization techniques in a single simulation engine, helping to improve system design concerning environmental issues. New methodologies are emerging, allowing to obtain optimal or sub-optimal solutions for these kind of systems without concessions on the necessary Validation and Verification processes. On the specific area of Smart Cities and Building energy management, different frameworks (e.g. NECADA) use this methodology to improve the sustainability in urban areas or buildings. This allows to combine IoT, Data Mining, Optimization, Simulation and other techniques through the formalized model.

« A formalized model respresents the structure and the behaviour of the model in a clear and unambiguous way. »

NECADA is a hybrid infrastructure that supports the execution of simulation models on cloud, cluster or desktop environments. The NECADA framework can optimize complete life cycles of buildings or urban areas, ranging from the design, building, actual use and the final rehabilitation and deconstruction processes, making the design of Net Zero Energy Buildings possible. The model implemented on NECADA considers the environmental directives and international rules in the design process (CEN TC 350) to optimize the behaviour of buildings or cities from the point of view of sustainability. It is defined using Specification and Description Language (SDL), a formal language from the Telecommunication Standardization Sector (from the UML family) with enhanced graphical capabilities, great expressiveness and easy to combine with other engineering languages. SDLPS, our software that understands SDL, performs an automatic simulation from this graphical representation of the model. No implementation is needed leading to a simplified verification, modification and expansion of the model. The simulation model definition is so simple that it can be accomplished and understood by all the members of the team, simplifying validation processes. Finally, co-simulation techniques are applied to state of the art calculus engines, like EnergyPlus, Trnsys, etc., using building information models to represent the complete building or area life cycles. Thanks to this architecture we can use NECADA to analyze thousands of different alternatives, using the power of super-computing or the precision of optimization algorithms to find the optimal solution for your building or urban area.

Intelligent support system for resilient port terminal

BMelian_MMoreno_postBy Belén Melián-Batista and J. Marcos Moreno-Vega, associate professors, University of La Laguna.

International freight plays an important role in the development and consolidation of important economic sectors such as trade, industry and tourism. The ports are critical infrastructures within multimodal transport networks that integrate maritime and terrestrial environments. They are also vulnerable to unforeseen events caused by nature or by deliberate or accidental human actions. The consequences range from the reduction of the productivity up to the temporary closure of the port.

In order to manage the risks and improve the robustness of the system, mechanisms that increase the resilience of these infrastructures are important. Resilience is the ability to withstand and recover intelligently from eventualities, facilitating the development of activities in the shortest possible time.

« The design, implementation and validation of a support system for the intelligent management to ensure the resilience of a port container terminal becomes a key issue »

Port operators must have tools to help them identify what unforeseen events can affect the port and what are their consequences. Moreover, they must design strategies to minimize its occurrence (proactive approach). If despite this fact, the event occurs, they should enable protocols and adopt decisions to repair the situation by minimizing the response time or the impact of such an event (reactive action).

For proactive management, the use of simulation tools with which to assess the impact of unexpected events on the overall system performance is mandatory. For reactive management, the design of intelligent replanning algorithms that allow us to restore the performance of the port to pre-incident levels is required.

Using statistics in the smart way: comparing impact factors and h-indices across disciplines

AJuan_RRuiz_post

By Rubén Ruiz, professor, Department of Applied Statistics, Operational Research and Quality, UPV; & Ángel A. Juan, associate professor, Computer Science Department, IN3 – UOC.

Data is everywhere. In the age of the internet we use data, computers, and statistics to analyze, measure, and compare everything: different strategies, policies, systems, processes, and even people. This allows us to be more informed and as a consequence, make better decisions. In a word, it makes us wiser than ever.

However, the ease of access to data and computers can also have some drawbacks if we are not cautious. Sometimes it is too easy to succumb to the temptation to compare apples with oranges, and consider the results of these comparisons as valid just because we are using some fancy math formulas, lots of Internet-based data sources, or powerful computers. Of course, extracting conclusions after comparing apples with oranges is a practice we should try to avoid.

In academics and research, this is what happens when we compare the impact factors (IF) of journals that belong to different disciplines. For instance, according to the ISI JCR 2013 Science Edition, the Reliability Engineering & System Safety journal has an IF of 2.048, and it belongs in the first quartile of its subject category ranking (Operations Research & Management Science).

« Extracting conclusions after comparing apples with oranges is a practice we should try to avoid »

However, whist having a similar IF (2.055), the Neuroscience Letters journal is located in the third quartile of its subject category (Neurosciences). To give an even more extreme example: Advanced Mathematics has an IF of 1.353 and belongs in the first quartile of its category (Mathematics), while Methods in Cell Biology, with an “equivalent” IF of 1.440, belongs in the fourth quartile of its category (Cell Biology).

Therefore, it is not reasonable to compare IFs belonging to different subject categories, at least not without applying some statistical adjustment to the index (e.g., normalizing by the IF range, etc.). Of course, it is even worse when somebody makes a direct comparison between two indices that use completely different criteria, e.g., the IF and the SCImago Journal Rank (SJR). Both are positive numbers used to establish journal rankings, but they are based on different mathematical formulas.

And what happens when comparing the h-indices of authors belonging to different disciplines? Recall that the h-index of an author is a given number “h” so that h of his or her Np papers have at least h citations each and the other (Np – h) papers have less than or equal to h citations each. Np is the number of papers published over ten years. So an author with an h-index of 10 has published at least 10 papers with at least 10 citations (or more).

Similarly to IF, as journals in some disciplines (e.g., Neurosciences) tend to receive many more citations than journals in other disciplines (e.g., Mathematics), it is obvious that a direct comparison of h-indices across disciplines will cause us the same problem as before. It is easy to argue that the h-index is a preferable measure when simply comparing the total papers published or the total received citations. However, it is not an indicator that one can use to compare between disciplines easily.

« Pure bibliometric indicators must be supplemented with an article-based assessment and peer-review of the past achievements of an individual »

As suggested in the paper by Ciriminna and Pagliaro1, it is good to use an h-index as a bibliometric indicator but its use should be made wisely, for example by doing an age-normalization or by considering the year of the first publication. Complementing the h-index with the SNIP (Source Normalized Index per Paper) or the aforementioned SJR would give a much richer picture.

In any case, there is a growing move towards the view that for greater accuracy, pure bibliometric indicators must be supplemented with an article-based assessment and peer-review of the past achievements of an individual. After all, it is obvious that no single easy to understand, easy to calculate metric can fully capture the rich and complex nature of the scholarly contribution of researchers to their disciplines, especially considering the rich and varied nature of these disciplines. Many forms and facets of scholarly achievement should be considered.

In summary, both the IF and the h-index are useful indicators, but they should be statistically adjusted and used with caution, especially when used to contrast journals or authors across disciplines. Despite this quite obvious conclusion, it is surprising to see how many times brilliant academics, researchers, and managers are still comparing apples with oranges. Shouldn’t we use statistics in a smarter way?

Further reading:

_____

1. Ciriminna, R. and Plagiaro, M., (2013). On the use of the h-index in evaluating chemical research. Chemistry Central, 7:132.

Bridging the gap between academy and industry: applied Operations Research/Analytics

AGarciaSanchez_post

By Álvaro García Sánchez, professor, Technical School of Industrial Engineering, Technical University of Madrid.

Optimization problems in real world systems are a huge challenge and are a source of enjoyment and effort for Operation Research/Analytics (OR/A) modelers as well of satisfaction when delivering good quality solutions. Nevertheless, other compelling challenges go along with devising a good solving strategy itself during a project where there is a customer.

A first challenge is to gain the credibility from the manager who is running a system that may me improved thought OR/A models. When there is a chance for modeling a problem and gaining effectiveness and efficiency in some way, prior to modeling itself, first we need to prove or inspire trust in the incumbent manager. Past failures, the gap between the practitioners and the academic worlds and so many other reasons may be stoppers.

The good news is that, once this barrier is overcome, the floodgates are opened for an ongoing improvement process getting more and more from increasingly comprehensive models, more valid, etc. In this sense, projects with a reduced scope can be a great approach for the managers to have a taste of what can be gained from OR/A and from modelers to learn more about the problem, and be in better conditions for having an estimate in terms of time and money of how the whole problem can be addressed.

« Projects with a reduced scope can be a great approach for the managers to have a taste of what can be gained from operations research/analytics »

Another challenge is getting to know the actual problem, so that it can be described in terms of a set of requirements. The modeler has to make a great effort to talk the manager’s language and distill what is relevant from what it is not. A typical issue appears when the client suggests some rules that are normally applied to come up with a solution to their everyday problem.

For example, since a particular process in a factory does not always require a night-scheduled worker, although nothing forbids scheduling that task during any other time of the day. This is not an actual requirement, but something that turns the problem into a tractable one for the scheduler. The modeler needs to know what an actual requirement is and what is not, not arrogantly but humbly, since a lot of knowledge is typically within the mind of some people and cannot be found in a structured document.

Other aspects may also be a source of endeavor, such as gathering and treating raw relevant data, integrating models with existing information systems, devising usable interfaces or managing out of scope demands during the development of the project.

Over the last years, the OR/A world has gained experiences in accomplishing successful projects not only in strictly mathematical terms, but also in the above mentioned key aspects. Thus, practitioners are in better conditions to rely in optimization and simulation tools and techniques for better managing their system. A time of closer cooperation and richer projects lies ahead of us.

Convergence of technologies in analytic databases

JGabaldon_post

By Jesús E. Gabaldón, analytics manager, Accenture Centre of Excellence.

In recent years, analytic databases have become crucial in many areas since they allow to provide the necessary workforce in sectors like social media, marketing, Internet of Things, among many others. For instance, everyday situations like sharing information in Twitter or Facebook would not be possible without this technology.

In this area, Massively Parallel Processing (MPP) as well as HDFS and MapReduce architectures have evolved separately most of the time, and both of them are still running on different tracks. However, each one is serving different purposes. Hadoop has a long record in high performance computing but it easily becomes an inaccurate and slow tool for certain tasks. There are some SQL-on-Hadoop databases but they are still immature and quite SQL unfriendly.

« Strengthening ties between universities and private sector is absolutely necessary for fulfilling the heavy demand of the business sector »

Several private and open source development lines are running in parallel. IBM, Pivotal and Apache Foundation are permanently extending the limits of performance in all directions: in-memory, MPP and Hadoop. However, MPP databases run a completely different strategy, probably the most powerful one for strongly demanding advanced analytic purposes. In any case, most of all are progressively relying on in-memory capabilities while caching more types of operations on progressively less expensive SSD supports.

MPP analytic databases are still less known than Hadoop but it is expected that they will turn soon into a strategic area, having a strong impact on highly process demanding tasks in sectors like digital marketing, banking and security. During the next years we will likely attend to a quick evolution of database and analytic technologies fuelled by the strong business demand.

However, this demand is not only requested on hardware performance but on technology development and integration, which often suffers from the scarcity of skilled and well trained professionals. Strengthening ties and establishing collaboration and partnership programs between universities and private sector is still absolutely necessary for fulfilling the heavy demand of the business sector.

Security and privacy in a smarter world

AEconomides_post

By Anastasios A. Economides, professor, University of Macedonia.

Internet of Things (IoT) is the worldwide Information and Communication Technologies (ICT) infrastructure that will support ubiquitous services among interacting humans, machines, data and applications. The proliferation of connected people, sensors, mobiles and applications is paving the way to IoT. It is expected that about 30 billion devices will be interconnected developing a market of around $ 2 trillion by 2020.

Most companies already use or plan to implement soon an IoT application in various IoT sectors, such as the following:

    • • Smart Healthcare & Wellbeing (e.g. Angel Sensor, Fitbit, Hexoskin, Intraway, Jawbone, Nymi, OnKöl Health Hub, Pebble, Philips Lifeline, Withings, Zebra MotionWorks).
    • • Smart Home & Building (e.g. Belkin, Nest, Neurio, Quirky, Sensorflare, SMA, SmartThings, Vivint, WallyHome, Withings, ZEN Thermostat).
    • • Smart City & Community (e.g. Bigbelly, Bitlock¸ FUKUSHIMA Wheel, Kiunsys, Placemeter, Silver Spring Networks, Waspmote).
    • • Smart Utilities (e.g. Enevo, Mayflower CMS, MeterNet, Osprey Informatics, Paradox, Trilliant).
    • • Smart Environmental Monitoring (e.g. FilesThruTheAir, Fruition Sciences, OnFarm, Semios, Topcon Precision Agriculture).
    • • Smart Car & Transportation (e.g. Audi, CarKnow, Connected Rail, Dash drive smart, Delphi Connect, Ericsson, Libelium, Logitrac, PowerFleet).
    • Smart Industry & Services (e.g. Argon Underground Mining Safety, Condeco Sense, DAQRI’s Smart Helmet, Numerex, Perch).

An amazing amount of confidential information will be collected, communicated, stored and processed by third parties. However, the consequences regarding security and privacy are unknown.

«Through IoT services, an amazing amount of confidential information will be collected, communicated, stored and processed by third parties»

Few companies admit to be prepared for tackling the IoT security challenges. A recent survey found that only 30% of IT professionals believe their company has the technology necessary to adequately evaluate the security of IoT devices. Also, 20% of the responders state that they have “no visibility” into current protection levels.

Security researchers from Hewlett-Packard performed a security audit of 10 popular Internet-connected devices: TVs, webcams, home thermostats, remote power outlets, sprinkler controllers, hubs for controlling multiple devices, door locks, home alarms, scales and garages. On average, they identified 25 vulnerabilities per device. Six out of the 10 devices that did not use encryption when downloading software updates. 70% of the devices used unencrypted network services and transmitted credentials in plain text. Also, 90% of the devices collected at least one piece of personal information via the device, the cloud, or its mobile application.

Last year, Team Cymru’s researchers identified over 300,000 home routers that have been attacked. This large scale attack was altering the Domain Name Server (DNS) settings of these compromised routers to redirect the victims’ DNS requests and subsequently replace the intended answers with Internet Protocol (IP) addresses and domains controlled by the attackers.

«Although many security officers believe their security is optimized and effective, in reality, this is not true»

Cisco just announced the results of its Security Capabilities Benchmark Survey among Chief Information Security Officers (CISOs) and Security Operations (SecOps) executives at 1700 companies in nine countries. Only 10% of Internet Explorer users run the latest version. Less than 50% of respondents use standard tools such as patching and configuration to help prevent security breaches and ensure that they are running the latest versions. Although many security officers believe their security is optimized and effective, in reality, this is not true.

It becomes apparent that society is not well prepared to face the security and privacy risks regarding IoT. Further security actions should be taken by ICT companies, businesses, public authorities as well individuals. Finally, I would like to invite you to a workshop that will discuss about those relevant issues in a few months: the IoT/CPS-Security 2015, which will be held in London from 8 to 12 June 2015.