An Experimental and Theoretical Comparison of Model Selection Methods

Machine Learning - We investigate the problem of model selection in the setting of supervised learning of boolean functions from independent random examples. More precisely, we compare methods for...     

Failure in internally pressurized bent tubes

The analysis and modeling of tube-hydroformed components is more complicated than that employed for sheet-metal panels, due to the lengthier process sequence and variable strain path—from flat-rolled sheet to tube; from straight tube to bent tube; and from bent tube to hydroformed component. These additional process steps make it difficult to determine whether post mortem analyses of tube failure during hydroforming can, and should, be conducted with the same tools and databases as used for simple stampings. To provide a partial answer, the properties of commercially fabricated welded straight tubes were evaluated using a free-expansion internal pressure test and compared with those of free-expansion internal pressure tests on bent tubes. The results demonstrated that the behavior of the bent tube was consistent with the mechanical properties of the as-received tube, provided due notice was accorded to the complex strain history of the bent tube. However, due to the strain-path changes occurring at the failure location, conventional approaches for monitoring strain history would yield (apparently) anomalous results.     

An evolutionary approach to pattern-based time series segmentation

Time series data, due to their numerical and continuous nature, are difficult to process, analyze, and mine. However, these tasks become easier when the data can be transformed into meaningful symbols. Most recent works on time series only address how to identify a given pattern from a time series and do not consider the problem of identifying a suitable set of time points for segmenting the time series in accordance with a given set of pattern templates (e.g., a set of technical patterns for stock analysis). However, the use of fixed-length segmentation is an oversimplified approach to this problem; hence, a dynamic approach (with high controllability) is preferable so that the time series can be segmented flexibly and effectively according to the needs of the users and the applications. In view of the fact that this segmentation problem is an optimization problem and evolutionary computation is an appropriate tool to solve it, we propose an evolutionary time series segmentation algorithm. This approach allows a sizeable set of pattern templates to be generated for mining or query. In addition, defining similarity between time series (or time series segments) is of fundamental importance in fitness computation. By identifying the perceptually important points directly from the time domain, time series segments and templates of different lengths can be compared and intuitive pattern matching can be carried out in an effective and efficient manner. Encouraging experimental results are reported from tests that segment both artificial time series generated from the combinations of pattern templates and the time series of selected Hong Kong stocks.     

Development of Downdrag on Piles and Pile Groups in Consolidating Soil

Development of pile settlement (downdrag) of piles constructed in consolidating soil may lead to serious pile foundation design problems. The investigation of downdrag has attracted far less attention than the study of dragload over the years. In this paper, several series of two-dimensional axisymmetric and three-dimensional numerical parametric analyses were conducted to study the behavior of single piles and piles in 3×3 and 5×5 pile groups in consolidating soil. Both elastic no-slip and elasto-plastic slip at the pile–soil interface were considered. For a single pile, the downdrag computed from the no-slip elastic analysis and from the analytical elastic solution was about 8–14 times larger than that computed from the elasto-plastic slip analysis. The softer the consolidating clay, the greater the difference between the no-slip elastic and the elasto-plastic slip analyses. For the 5×5 pile group at 2.5 diameter spacing, the maximum downdrag of the center, inner, and corner piles was, respectively, 63, 68, and 79% of the maximum downdrag of the single pile. The reduction of downdrag inside the pile group is attributed to the shielding effects on the inner piles by the outer piles. The relative reduction in downdrag (Wr) in the 5×5 pile group increases with an increase in the relative bearing stiffness ratio (Eb/Ec), depending on the pile location in the group. Compared with the relative reduction in dragload (Pr), Wr at the corner pile is less affected by the group interaction for a given surcharge load. This suggests that the use of sacrificing piles outside the pile group will be more effective on Pr than on Wr. Based on the three cases studied, the larger the number of piles in a group, the greater the shielding effects on Wr. Relatively speaking, Wr is more sensitive to the total number of piles than to the pile spacing within a pile group.     

A Personalized Defection Detection and Prevention Procedure based on the Self-Organizing Map and Association Rule Mining: Applied to Online Game Site

Customer retention is an increasinglypressing issue in today's competitiveenvironment. This paper proposes a personalizeddefection detection and prevention procedurebased on the observation that potentialdefectors have a tendency to take a couple ofmonths or weeks to gradually change theirbehaviour (i.e., trim-out their usage volume)before their eventual withdrawal. For thispurpose, we suggest a SOM (Self-Organizing Map)based procedure to determine the possiblestates of customer behaviour from pastbehaviour data. Based on this staterepresentation, potential defectors aredetected by comparing their monitoredtrajectories of behaviour states with frequentand confident trajectories of past defectors.Also, the proposed procedure is extended toprevent the defection of potential defectors byrecommending the desirable behaviour state forthe next period so as to lower the likelihoodof defection. For the evaluation of theproposed procedure, a case study has beenconducted for a Korean online game site. Theresult demonstrates that the proposed procedureis effective for defection prevention andefficiently detects potential defectors withoutdeterioration of prediction accuracy whencompared to that of the MLP (Multi-LayerPerceptron) neural networks.     

Analytical and Approximate Variance of Total Project Cost

The statistical variance of total project cost is usually estimated by means of Monte Carlo simulation on the assumption that exact analytic approaches are too difficult. This paper tests that assumption and shows that, contrary to expectations, the analytic solution is relatively straightforward. It is also shown that the coefficient of variation is unaffected by the size (floor area) of the project when using standardized component costs. A case study is provided in which actual component costs are analyzed to obtain the required total cost variance. The results confirm previous work in showing that the approximation of the second moment (variance) under the assumption of independence considerably underestimates the exact value. The analysis then continues to examine the effects of professional judgment, and, with the simulated data used, the approximation is shown to be reasonably accurate—the professional judgment absorbing most of the intercorrelations involved. An example is also given in which the component unit quantities are priced by their average unit costs and which again shows the approximation to be close to the true value. Finally, this is extended to show how the exact total project cost variances may be obtained for each project.     

Predicting the Academic Performance of Construction Engineering Students by Teaching and Learning Approaches: Case Study

There are various teaching approaches that instructors may adopt in their quest to teach effectively, and students can choose from a range of learning approaches to help them achieve good grades. This study investigates the effectiveness of personal learning approaches adopted by undergraduates and the teaching approaches employed by instructors in the context of construction engineering courses. The research questions are as follows. (1)?What are the students’ learning approaches that lead to better academic achievement? (2)?What are the instructors’ teaching approaches that give rise to better academic achievement? The study also aims to construct a model to predict a student’s likely academic performance in a construction engineering course. The research design is a self-administered survey. Using a structured questionnaire consisting of questions relating to learning strategies and teaching approaches, data were collected from undergraduates who majored in construction engineering. Statistical analyses undertaken include Pearson correlation and multiple linear regression modeling. Pearson correlation analysis shows that both the Growing Teaching approach and the students’ Achieving Motive learning approach are significantly correlated with academic performance. A robust grade prediction model was developed, whereby a student’s grade in a construction engineering course may be predicted using one teaching approach (Growing Teaching) and three learning approaches (Achieving Motive, Achieving Strategy, and Deep Motive). To help students obtain higher grades in a construction engineering course, instructors should adopt the Growing Teaching approach, and students should adopt the Achieving Motive learning approach. Because students who adopt the Deep Motive learning approach are penalized with lower grades, it is recommended that instructors modify their teaching methods and approaches so that students do not take shortcuts in their learning without deeper regard for higher-order learning outcomes.