Service Analytics
Permanent URI for this collection
Browse
Recent Submissions
1 - 7 of 7
-
ItemEstablishing an Extendable Benchmarking Framework for E-Fulfillment( 2020-01-07)The growth in attended home deliveries motivates research in prescriptive analytics for e-fulfillment. Introducing new analytics solutions, for instance, for vehicle routing or revenue management, requires simulation-based benchmarking and analyses on relevant problem scenarios. Unfortunately, creating the required systems induces high overhead for analytics researchers. This paper introduces the simulation-based benchmarking framework SiLFul, which aims to support scientific rigor and practical relevance of research by reducing this overhead. It provides a toolbox of approaches, a modular and extendable architecture, and a comprehensive, application-related data model. Thereby, it facilitates controllable analyses and transparent and replicable research. Moreover, we propose a research process that leverages the framework for evaluating analytics and allows continuous development of the framework as a community effort.
-
ItemTowards Predictive Part Quality and Predictive Maintenance in Industrial Machining - A Data-Driven Approach( 2020-01-07)Programs such as Industry 4.0 and Internet of Things contain the promise of "intelligent production" with "smart services". In fact, great advances have already been made in sensor technology and machine connectivity. Production plants continuously generate and communicate large amounts of data and have become "cyber-physical systems". However, the task of gaining knowledge from these large amounts of data is still challenging. Data generated by numerical control (NC) and programmable logic controllers (NC) comes in a raw format that doesn’t allow the application of analytical methods directly. Extensive preprocessing and feature engineering has to be applied to structure this data for further analysis. An important application is the timely detection of deviations in the production process which allows immediate reactions and adjustments of production parameters or indicates the necessity of a predictive maintenance action. In our research, we aimed at the identification of special deviant behavior of a grinding machine based on NC data. One finding wast the distinguishing the warm-up program from regular production and the other to recognize imprecise identification of the grinding process window. Both tasks could be solved with extensive preprocessing of the raw data, appropriate feature extraction and feature reduction, and the subsequent application of a clustering algorithm.
-
ItemOptimizing the Cloud Data Center Availability Empowered by Surrogate Models( 2020-01-07)Making data centers highly available remains a challenge that must be considered since the design phase. The problem is selecting the right strategies and components for achieving this goal given a limited investment. Furthermore, data center designers currently lack reliable specialized tools to accomplish this task. In this paper, we disclose a formal method that chooses the components and strategies that optimize the availability of a data center while considering a given budget as a constraint. For that, we make use of stochastic models to represent a cloud data center infrastructure based on the TIA-942 standard. In order to improve the computational cost incurred to solve this optimization problem, we employ surrogate models to handle the complexity of the stochastic models. In this work, we use a Gaussian process to produce a surrogate model for a cloud data center infrastructure and we use three derivative-free optimization algorithms to explore the search space and to find optimal solutions. From the results, we observe that the Differential Evolution (DE) algorithm outperforms the other tested algorithms, since it achieves higher availability with a fair usage of the budget.
-
ItemTowards More Robust Uplift Modeling for Churn Prevention in the Presence of Negatively Correlated Estimation Errors( 2020-01-07)The subscription economy is rapidly growing, boosting the importance of churn prevention. However, current true lift models often lead to poor outcomes in churn prevention campaigns. A vital problem seems to lie in instable estimations due to dynamic surrounding parameters such as price increases, product migrations, tariff launches of a competitor, or other events with uncertain consequences. The crucial challenge therefore is to make churn prevention measures more reliable in the presence of game-changing events. In this paper, we assume such events to be spatially finite in feature space, an assumption which leads to particularly bad churn prevention results if the selected customers lump in an affected region of the feature space. We then introduce novel methods which trade off uplift for reduced similarity in feature space when selecting customers for churn prevention campaigns and show that these methods can improve the robustness of uplift modeling.
-
ItemApplying Optimal Weight Combination in Hybrid Recommender Systems( 2020-01-07)We propose a method for learning weighting schemes in weighted hybrid recommender systems (RS) that is based on statistical forecast and portfolio theory. An RS predicts the future preference of a set of items for a user, and recommends the top items. A hybrid RS combines individual RS in making the predictions. To determine the weighting of individual RS, we learn so-called optimal weights from the covariance matrix of available error data of individual RS that minimize the error of a combined RS. We test the method on the well-known MovieLens 1M dataset, and, contrary to the “forecast combination puzzle”, stating that a simple average (SA) weighting typically outperforms learned weights, the out-of-sample results show that the learned weights consistently outperform the individually best RS as well as an SA combination.
-
ItemA Next Click Recommender System for Web-based Service Analytics with Context-aware LSTMs( 2020-01-07)Software companies that offer web-based services instead of local installations can record the user’s interactions with the system from a distance. This data can be analyzed and subsequently improved or extended. A recommender system that guides users through a business process by suggesting next clicks can help to improve user satisfaction, and hence service quality and can reduce support costs. We present a technique for a next click recommender system. Our approach is adapted from the predictive process monitoring domain that is based on long short-term memory (LSTM) neural networks. We compare three different configurations of the LSTM technique: LSTM without context, LSTM with context, and LSTM with embedded context. The technique was evaluated with a real-life data set from a financial software provider. We used a hidden Markov model (HMM) as the baseline. The configuration LSTM with embedded context achieved a significantly higher accuracy and the lowest standard deviation.
-
ItemIntroduction to the Minitrack on Service Analytics( 2020-01-07)