A statement of the purpose for your essay and a brief outline of how you will discuss the selected article (one or two paragraphs). Make sure to identify the article being reviewed.
Background of the Topic
The aim of this essay is to discuss and extensively elaborate on the proceeds of the 11th through the 13th international conferences that were conducted on Business Process Management. The paper discussed has a direction relevance to business process management. The selection conference article for this essay is Mining Resource-Scheduling Protocols. The competition for the scarce commodities and capacity of the providers of the various services cuts across the board. Just like the healthcare, financial and telecommunication sectors, as found in them service processes, the competition is conspicuous (Guharoy et al., 2017). This leads to the need of a performance analysis which mostly aims at the moment when consumers are mostly delayed before service even though such a wait time may not be extensively and comprehensively explained by the loads that are put on the service providers.
The focus of this paper is on automatically learning resource decisions that are borrowed from events. A hypothesis made in the paper is that queuing information play a fundamental role as an an important aspect in mining for example protocol and thus the queuing perspective of customers is adopted in this paper in the mining process (Guharoy et al., 2017).
Service processes may be perceived as a unique business process case where the consumer also known as the customers is in a continuous competition for the limited capacity of the service providers that are available. Service providers may be found for example in medical, telecommunication and even financial sectors (Senderovich, Weidlich, Gal & Mandelbaum, 2014).
One of the features of service processes is the high capacities of the requests of services that are a subject to great variation over the time as well as high uncertainty levels in the quantity of incoming demand. Hence, in a bid to guarantee a successful performance of a service process, for example reducing the waiting tine for services by customers, one has to determine an effective level of resource capacity (Senderovich, 2017).
The paper addressed different operational process mining techniques that handle the prediction of the times for waiting based on the features of the service requests. The work began with a notice that the processing delays may not be elaborately expanded using the process load and instead the delays as well emanate from the service providers themselves (Sun et al., 2016). The observation was that the service providers have a symmetric role to play when it comes to influencing the process performance in comparison with the clients. The service provider’s teds to abide by a resource scheduling protocol which enables a prediction of the subsequent responsibility of a service provider, from a given set of feasible tasks hence offering a more complete performance analysis as well as an enhancement in the accuracy of the prediction of waiting time (Guharoy et al., 2017).
The second section of the paper is regarding the background of the topic which entails among other discussions service processes as interacting processes, resource scheduling protocols, delays in service processes and service logs. The problem statement of the paper was mining resource scheduling protocols. These are the protocols that are sued in the determination of the modalities of resource allocation to queues of customers and entail the selection of a certain type of customer (Guharoy et al., 2017).
Experimental method of research was adopted with the aim of meeting the objectives of the paper. A set of large scale real world data was used and the initial step involved the provision of the details regarding the experimental set up as well as the data to be used. A report and thereafter a discussion were done regarding the prediction results.
The data used for the experiment was extracted from one of the call centers in an Israeli telecommunication company and was collected, thereafter kept in the Technion laboratory. Israeli Telecommunication Company handles to the tune of 5000 service requests on a daily basis and is operation by an average of 700 agents on a weekday and 300 during the weekends. The company offers numerous service types among the quite often being Technical, Private, and Commercial as well as Content Internet (Biason et al., 2017). The paper gave focus to Private Service that was in charge of regular, low and VIP priorities. Three months of data was selected for empirical evaluation to act as the service logs for the experiment from the first date of January to the last date of March, 2008. The information collected contained the events of the service log of a customer alongside the log service log of a resource.
The experiment was set up in such a way that the method applied to mine a given scheduling protocol was the control variable (Senderovich, Weidlich, Gal & Mandelbaum, 2015). The methods were mainly divided into two: queuing heuristics which depended on the control theory in heavy traffic and techniques for data mining which were based on the characteristic vectors which could be dependent on queuing information. Precision was responding or uncontrolled variable in the experiment. The expector predictor error was actually a fraction between zero and one as 0-1 was taken into account as the loss function and thereby the precision rate was 1-EPE, which means the complimentary of the prediction error (Thomas, Venkateswaran, Singh & Krishnamoorthy, 2014).
Findings and Discussion
The experiment was composed of four scenarios which were corresponding to each of the four ranks of queuing information. While Rank I was the baseline rank and had the feature vector be inclusive of the resource skill collection only excluding further queuing information, ranks II took into consideration the lengths of the queues of the 3 queues used (low-priority, VIP and regular) as the main extra features (Guharoy et al., 2017). The length of the queue is replaced with the head of line waiting time for very queue in rank III as rank IV encompasses aforementioned scenarios among the skill, head of line waiting time as well as length of the queue (Fu et al., 2015).
Running the experiment first included reducing the allocation decision as well as the feature vectors. For every experiment, the allocation decisions were randomly divided into two groups: a test set that took 25% and training set that took 75% of the allocation which is a common practice when carrying out an assessment of statistical model (Luo et al., 2018). The controlled variables were changed during each iteration and the process of data set division was repeated and the experiment runs 10 times.
The findings of the experiment were as recorded in figure 1 below that plotted the obtained precision for all the six methods used: two for queuing heuristics and four for the purposes of data mining (Guharoy et al., 2017). There were four results for each of the methods for data mining, one for every type of information queuing.
Figure 1: Precision rates of the discovered protocols: Extracted from (Guharoy et al., 2017)
As could be noted from the results, all the data mining algorithms generated the same precision for the baseline scenario when resource skill group was the only available information. There was no improvement into the LDA in scenarios II to IV beyond the baseline scenario upon the introduction of queuing. There was an 8% increase in MLR upon taking into consideration the length of the queue in the prediction (Guharoy et al., 2017).
Nonetheless, this was by far and large inferior in comparison with the decision trees, Longest-Queue-First heuristics as well as random forests that obtained 83%, 81% and 84% precision in that order. Taking into consideration the impact of head of line waiting time, there was no improvement in any of the linear classification yet precisions of 84%, 79% and 86% respectively were obtained for decision trees, Most-Delayed-First heuristic and random forests. There was recorded a slight improvement in the decision trees and random forests upon the consideration of all the queuing information types (Kobusi?ska et al., 2018).
With regard to discussion on the obtained results, the first observation made is that the techniques of linear data mining generated relatively low values of precision. This could be elaborated by their grip presumptions for example LDA makes an assumption of the Gaussian densities of the characteristic vector that are aligned to the decisions (Shu et al., 2016). As a result, the linear techniques put on a computation effort that is relatively low and should the respective assumptions remain anything to go by, a precise classification is offered (Kumar, Ranjan, Ramaswami & Tripathy, 2017). Nonetheless, owing to the assumptions they make, the techniques cannot be applied to any of the four scenarios and hence resulting in poor performance of the used data set.
On the contrary, there are no assumptions imposed by the random forests and the decision trees on the characteristics vector and thus greater robustness to different spreads of such vectors. The tree techniques are mainly pegged on the greedy algorithms, making them least likely to converge at an optimal division of the space of the feature vector. Even with this limitation, the tree based methods generated the best prediction results in the comparative analysis besides enabling the decrypting of sophisticated scheduling protocols from event information (Tesfaye, Zhang, Zheng & Shiferaw, 2016).
It is of great importance to have a comprehensive understanding of the resource scheduling protocols which are congruent with the customers with service providers for the analysis of the performance of the service processes and estimation of delays processing. Going by the fact that in most cases, information systems used for tracing the customers activities and the behavior of resources support service processes, the discovery of such protocols is highly encouraged to be from event data. Going by the hypothesis of the paper that queuing information plays an important role as a main element in the mining scheduling protocols, two main and specific methods were presented and elaborately discussed in the paper. The first one involved illustrating the use of classification techniques from data mining when queuing information is concluded in the explanatory features.
The second aspect involved a proposal of the heuristics that comes from queuing guess, exploring mainly the existing condition of a given setup. Both two technique types were tested with the aid of a set of large world real data from the telecommunication sector. The findings from the experiment illustrated that data mining using random forests and decision trees can extract predictors for decision scheduling to the tune of 88% precision. Still, queuing heuristics was also established to exhibit good performance that could get to the levels of 81% precision.
A conclusion could thus be made that it is possible to attain high prediction precision already using online techniques that do not need a phase of elementary learning on historic information. Section of the future task of the paper was an aim of coming up with and examining queuing heuristics which are enriched with a small features sett derived from historical data and hence enabling for another prediction precision and effort of trade-off.
Biason, A., Pielli, C., Rossi, M., Zanella, A., Zordan, D., Kelly, M. and Zorzi, M., 2017. EC-CENTRIC: an energy-and context-centric perspective on IoT systems and protocol design. IEEE Access, 5, pp.6894-6908
Fu, T.Z., Ding, J., Ma, R.T., Winslett, M., Yang, Y. and Zhang, Z., 2015, June. DRS: dynamic resource scheduling for real-time analytics over fast streams. In Distributed Computing Systems (ICDCS), 2015 IEEE 35th International Conference on (pp. 411-420). IEEE
Guharoy, R., Sur, S., Rakshit, S., Kumar, S., Ahmed, A., Chakborty, S., Dutta, S. and Srivastava, M., 2017, August. A theoretical and detail approach on grid computing a review on grid computing applications. In Industrial Automation and Electromechanical Engineering Conference (IEMECON), 2017 8th Annual (pp. 142-146). IEEE
Kobusi?ska, A., Leung, C., Hsu, C.H., Raghavendra, S. and Chang, V., 2018. Emerging trends, issues and challenges in Internet of Things, Big Data and cloud computing
Kumar, S., Ranjan, P., Ramaswami, R. and Tripathy, M.R., 2017. Resource efficient clustering and next hop knowledge based routing in multiple heterogeneous wireless sensor networks. International Journal of Grid and High Performance Computing (IJGHPC), 9(2), pp.1-20
Luo, F., Dong, Z.Y., Xu, Z., Kong, W. and Wang, F., 2018. Distributed residential energy resource scheduling with renewable uncertainties. IET Generation, Transmission & Distribution, 12(11), pp.2770-2777
Senderovich, A., 2017. Queue Mining: Service Perspectives in Process Mining. In BPM (Demos)
Senderovich, A., Weidlich, M., Gal, A. and Mandelbaum, A., 2014, September. Mining resource scheduling protocols. In International Conference on Business Process Management(pp. 200-216). Springer, Cham
Senderovich, A., Weidlich, M., Gal, A. and Mandelbaum, A., 2015. Queue mining for delay prediction in multi-class service processes. Information Systems, 53, pp.278-295
Shu, Z., Wan, J., Zhang, D. and Li, D., 2016. Cloud-integrated cyber-physical systems for complex industrial applications. Mobile Networks and Applications, 21(5), pp.865-878
Sun, X., Hu, C., Yang, R., Garraghan, P., Wo, T., Xu, J., Zhu, J. and Li, C., 2018. ROSE: Cluster Resource Scheduling via Speculative Over-subscription
Tesfaye, A., Zhang, J.H., Zheng, D.H. and Shiferaw, D., 2016. Short-term wind power forecasting using artificial neural networks for resource scheduling in microgrids. International Journal of Science and Engineering Applications (IJSEA), 5(3)
Thomas, A., Venkateswaran, J., Singh, G. and Krishnamoorthy, M., 2014. A resource constrained scheduling problem with multiple independent producers and a single linking constraint: A coal supply chain example. European Journal of Operational Research, 236(3), pp.946-956