Kernel-Based Ranking. Methods for Learning and Performance Estimation

Machine learning provides tools for automated construction of predictive models in data intensive areas of engineering and science. The family of regularized kernel methods have in the recent years become one of the mainstream approaches to machine learning, due to a number of advantages the methods share. The approach provides theoretically well-founded solutions to the problems of under- and overfitting, allows learning from structured data, and has been empirically demonstrated to yield high predictive performance on a wide range of application domains. Historically, the problems of classification and regression have gained the majority of attention in the field. In this thesis we focus on another type of learning problem, that of learning to rank. In learning to rank, the aim is from a set of past observations to learn a ranking function that can order new objects according to how well they match some underlying criterion of goodness. As an important special case of the setting, we can recover the bipartite ranking problem, corresponding to maximizing the area under the ROC curve (AUC) in binary classification. Ranking applications appear in a large variety of settings, examples encountered in this thesis include document retrieval in web search, recommender systems, information extraction and automated parsing of natural language. We consider the pairwise approach to learning to rank, where ranking models are learned by minimizing the expected probability of ranking any two randomly drawn test examples incorrectly. The development of computationally efficient kernel methods, based on this approach, has in the past proven to be challenging. Moreover, it is not clear what techniques for estimating the predictive performance of learned models are the most reliable in the ranking setting, and how the techniques can be implemented efficiently. The contributions of this thesis are as follows. First, we develop RankRLS, a computationally efficient kernel method for learning to rank, that is based on minimizing a regularized pairwise least-squares loss. In addition to training methods, we introduce a variety of algorithms for tasks such as model selection, multi-output learning, and cross-validation, based on computational shortcuts from matrix algebra. Second, we improve the fastest known training method for the linear version of the RankSVM algorithm, which is one of the most well established methods for learning to rank. Third, we study the combination of the empirical kernel map and reduced set approximation, which allows the large-scale training of kernel machines using linear solvers, and propose computationally efficient solutions to cross-validation when using the approach. Next, we explore the problem of reliable cross-validation when using AUC as a performance criterion, through an extensive simulation study. We demonstrate that the proposed leave-pair-out cross-validation approach leads to more reliable performance estimation than commonly used alternative approaches. Finally, we present a case study on applying machine learning to information extraction from biomedical literature, which combines several of the approaches considered in the thesis. The thesis is divided into two parts. Part I provides the background for the research work and summarizes the most central results, Part II consists of the five original research articles that are the main contribution of this thesis.

APROS model validation in case of fast transients in a boiling water reactor

This Master´s thesis investigates the performance of the Olkiluoto 1 and 2 APROS model in case of fast transients. The thesis includes a general description of the Olkiluoto 1 and 2 nuclear power plants and of the most important safety systems. The theoretical background of the APROS code as well as the scope and the content of the Olkiluoto 1 and 2 APROS model are also described. The event sequences of the anticipated operation transients considered in the thesis are presented in detail as they will form the basis for the analysis of the APROS calculation results. The calculated fast operational transient situations comprise loss-of-load cases and two cases related to a inadvertent closure of one main steam isolation valve. As part of the thesis work, the inaccurate initial data values found in the original 1-D reactor core model were corrected. The input data needed for the creation of a more accurate 3-D core model were defined. The analysis of the APROS calculation results showed that while the main results were in good accordance with the measured plant data, also differences were detected. These differences were found to be caused by deficiencies and uncertainties related to the calculation model. According to the results the reactor core and the feedwater systems cause most of the differences between the calculated and measured values. Based on these findings, it will be possible to develop the APROS model further to make it a reliable and accurate tool for the analysis of the operational transients and possible plant modifications.

Performance of an anaerobic-aerobic reactor and kinetic study of organic matter removal of cattle slaughterhouse effluent

This paper sought to evaluate the behavior of an upflow Anaerobic-Aerobic Fixed Bed Reactor (AAFBR) in the treatment of cattle slaughterhouse effluent and determine apparent kinetic constants of the organic matter removal. The AAFBR was operated with no recirculation (Phase I) and with 50% of effluent recirculation (Phase II), with θ of 11h and 8h. In terms of pH, bicarbonate alkalinity and volatile acids, the results indicated the reactor ability to maintain favorable conditions for the biological processes involved in the organic matter removal in both operational phases. The average removal efficiencies of organic matter along the reactor height, expressed in terms of raw COD, were 49% and 68% in Phase I and 54% and 86% in Phase II for θ of 11h and 8h, respectively. The results of the filtered COD indicated removal efficiency of 52% and k = 0.0857h-1 to θ of 11h and 42% and k = 0.0880h-1 to θ of 8h in the Phase I. In Phase II, the removal efficiencies were 59% and 51% to θ of 11h and 8h, with k = 0.1238h-1 and k = 0.1075 h-1, respectively. The first order kinetic model showed good adjustment and described adequately the kinetics of organic matter removal for θ of 11h, with r² equal to 0.9734 and 0.9591 to the Phases I and II, respectively.

Hydrological simulation using SWAT model in headwater basin in Southeast Brazil

Hydrological models are important tools that have been used in water resource planning and management. Thus, the aim of this work was to calibrate and validate in a daily time scale, the SWAT model (Soil and Water Assessment Tool) to the watershed of the Galo creek , located in Espírito Santo State. To conduct the study we used georeferenced maps of relief, soil type and use, in addition to historical daily time series of basin climate and flow. In modeling were used time series corresponding to the periods Jan 1, 1995 to Dec 31, 2000 and Jan 1, 2001 to Dec 20, 2003 for calibration and validation, respectively. Model performance evaluation was done using the Nash-Sutcliffe coefficient (E NS) and the percentage of bias (P BIAS). SWAT evaluation was also done in the simulation of the following hydrological variables: maximum and minimum annual daily flowsand minimum reference flows, Q90 and Q95, based on mean absolute error. E NS and P BIAS were, respectively, 0.65 and 7.2% and 0.70 and 14.1%, for calibration and validation, indicating a satisfactory performance for the model. SWAT adequately simulated minimum annual daily flow and the reference flows, Q90 and Q95; it was not suitable in the simulation of maximum annual daily flows.

Use of interactive performance optimization for identifying the ideal profile of swine finishing producers

The present research aimed to develop a modeling capable of identifying the ideal profile of swine finishing producers using the interactive performance optimization, which began by verifying qualitative the criteria considered most relevant to the decision-making, generating a closed structured diagnosis that covers the socioeconomic aspects about the activity, until the design of a mathematical model able to translate the data obtained in quantitative information. For the verification, it was proposed a practical study for a universe of 120 members of a cooperative in the state of Rio Grande do Sul, Brazil. The results showed that, from the application and the definition of the ideal profile, it was possible to verify that 82 producers are in the group of those who have obtained a "Good" performance, and to 44 the result is in the range between 86% to 90% from the ideal, which means that most have short or medium-term conditions to evolve their status for the considered "Very Good", where only 12.5% of the producers are currently.

Fuzzy modeling of broiler performance, raised from 1 to 21 days, subject to heat stress

ABSTRACT Given the need to obtain systems to better control broiler production environment, we performed an experiment with broilers from 1 to 21 days, which were submitted to different intensities and air temperature durations in conditioned wind tunnels and the results were used for validation of afuzzy model. The model was developed using as input variables: duration of heat stress (days), dry bulb air temperature (°C) and as output variable: feed intake (g) weight gain (g) and feed conversion (g.g-1). The inference method used was Mamdani, 20 rules have been prepared and the defuzzification technique used was the Center of Gravity. A satisfactory efficiency in determining productive responses is evidenced in the results obtained in the model simulation, when compared with the experimental data, where R2 values ​​calculated for feed intake, weight gain and feed conversion were 0.998, 0.981 and 0.980, respectively.

Performance of small wind turbines: simulation of electricity supply to loads connected to the public or isolated grid

ABSTRACT The successful in the implementation of wind turbines depends on several factors, including: the wind resource at the installation site, the equipment used, project acquisition and operational costs. In this paper, the production of electricity from two small wind turbines was compared through simulation using the computer software HOMER - a national model of 6kW and an imported one of 5kW. The wind resources in three different cities were considered: Campinas (SP/BR), Cubatão (São Paulo/BR) and Roscoe (Texas/ USA). A wind power system connected to the grid and a wind isolated system - batteries were evaluated. The results showed that the energy cost ($/kWh) is strongly dependent on the windmill characteristics and local wind resource. Regarding the isolated wind system – batteries, the full supply guarantee to the simulated electrical load is only achieved with a battery bank with many units and high number of wind turbines, due to the intermittency of wind power.

Chasing performance persistence of hedge funds

The thesis examines the performance persistence of hedge funds using complement methodologies (namely cross-sectional regressions, quantile portfolio analysis and Spearman rank correlation test). In addition, six performance ranking metrics and six different combinations of selection and holding periods are compared. The data is gathered from HFI and Tremont databases covering over 14,000 hedge funds and time horizon is set from January 1996 to December 2007. The results suggest that there definitely exists performance persistence among hedge funds and the strength and existence of persistence vary among fund styles. The persistence depends on the metrics and combination of selection and prediction period applied. According to the results, the combination of 36-month selection and holding period outperforms other five period combinations in capturing performance persistence within the sample. Furthermore, model-free performance metrics capture persistence more sensitively than model-specific metrics. The study is the first one ever to use MVR as a performance ranking metric, and surprisingly MVR is more sensitive to detect persistence than other performance metrics employed.

Mutual Fund Performance and the Market Timing Ability - Finnish Evidence

The purpose of this thesis is to examine the performance of Finnish equity funds and their market timing ability. Fund performance is evaluated by using annual returns and various risk-adjusted measures, including Sharpe ratio, DDSR, SKASR, Treynor ratio and Jensen’s alpha, whereas portfolio manager’s timing ability is examined with Treynor-Mazuy model and Henriksson-Merton model. The data is collected from the Finnish fund market during the sample period from January 1997 to February 2010. Results show that Finnish equity funds have been able to outperform the market return on a risk-adjusted basis, but these results are influenced heavily by the exceptionally good performance during the IT-bubble. Market timing models show that fund managers have been, to some degree, able to time the market but not a single fund have been able to possess security selection ability and market timing ability simultaneously.

Marketing-Sales Cooperation and its Relation on Direct Sales Performance: an Example of a Medium-sized Oil Company

The study touches upon marketing-sales departments’ cooperation and investigates marketing-sales cooperative model within the case company. So that research increases understanding of linkages between Marketing and Sales departments with an illustrative example of Russian medium-sized oil company (LLC Neste St. Petersburg), the subsidiary of Finnish-based Neste Oil. The empirical study is done from marketing and sales perspectives. And for sales main attention was brought to direct sales, both B2B and B2C. Research considers all five domains of cooperation, and among others, study reveals the attitude towards external (market) and internal (product) knowledge, and its mutual use by marketing and sales managers. A qualitative research method, participant observations, and in-depth interviews with upper-management made it possible to explore all facets of joint work. Moreover, research responses the changes in a model of cooperation between marketing and sales when moving from medium size to large company.

Generic Life Cycle Cost Model and Cost-effective Solutions

Life cycle costing (LCC) practices are spreading from military and construction sectors to wider area of industries. Suppliers as well as customers are demanding comprehensive cost knowledge that includes all relevant cost elements through the life cycle of products. The problem of total cost visibility is being acknowledged and the performance of suppliers is evaluated not just by low acquisition costs of their products, but by total value provided through the life time of their offerings. The main purpose of this thesis is to provide better understanding of product cost structure to the case company. Moreover, comprehensive theoretical body serves as a guideline or methodology for further LCC process. Research includes the constructive analysis of LCC related concepts and features as well as overview of life cycle support services in manufacturing industry. The case study aims to review the existing LCC practices within the case company and provide suggestions for improvements. It includes identification of most relevant life cycle cost elements, development of cost breakdown structure and generic cost model for data collection. Moreover, certain cost-effective suggestions are provided as well. This research should support decision making processes, assessment of economic viability of products, financial planning, sales and other processes within the case company.

A multibody dynamic model of the cross-country ski-skating technique

The objective of this thesis is the development of a multibody dynamic model matching the observed movements of the lower limb of a skier performing the skating technique in cross-country style. During the construction of this model, the formulation of the equation of motion was made using the Euler - Lagrange approach with multipliers applied to a multibody system in three dimensions. The description of the lower limb of the skate skier and the ski was completed by employing three bodies, one representing the ski, and two representing the natural movements of the leg of the skier. The resultant system has 13 joint constraints due to the interconnection of the bodies, and four prescribed kinematic constraints to account for the movements of the leg, leaving the amount of degrees of freedom equal to one. The push-off force exerted by the skate skier was taken directly from measurements made on-site in the ski tunnel at the Vuokatti facilities (Finland) and was input into the model as a continuous function. Then, the resultant velocities and movement of the ski, center of mass of the skier, and variation of the skating angle were studied to understand the response of the model to the variation of important parameters of the skate technique. This allowed a comparison of the model results with the real movement of the skier. Further developments can be made to this model to better approximate the results to the real movement of the leg. One can achieve this by changing the constraints to include the behavior of the real leg joints and muscle actuation. As mentioned in the introduction of this thesis, a multibody dynamic model can be used to provide relevant information to ski designers and to obtain optimized results of the given variables, which athletes can use to improve their performance.

Practical Framework for Monitoring the Performance of Agile IT Projects in an Output-Based Outsourcing Context

Over the past decade, organizations worldwide have begun to widely adopt agile software development practices, which offer greater flexibility to frequently changing business requirements, better cost effectiveness due to minimization of waste, faster time-to-market, and closer collaboration between business and IT. At the same time, IT services are continuing to be increasingly outsourced to third parties providing the organizations with the ability to focus on their core capabilities as well as to take advantage of better demand scalability, access to specialized skills, and cost benefits. An output-based pricing model, where the customers pay directly for the functionality that was delivered rather than the effort spent, is quickly becoming a new trend in IT outsourcing allowing to transfer the risk away from the customer while at the same time offering much better incentives for the supplier to optimize processes and improve efficiency, and consequently producing a true win-win outcome. Despite the widespread adoption of both agile practices and output-based outsourcing, there is little formal research available on how the two can be effectively combined in practice. Moreover, little practical guidance exists on how companies can measure the performance of their agile projects, which are being delivered in an output-based outsourced environment. This research attempted to shed light on this issue by developing a practical project monitoring framework which may be readily applied by organizations to monitor the performance of agile projects in an output-based outsourcing context, thus taking advantage of the combined benefits of such an arrangement Modified from action research approach, this research was divided into two cycles, each consisting of the Identification, Analysis, Verification, and Conclusion phases. During Cycle 1, a list of six Key Performance Indicators (KPIs) was proposed and accepted by the professionals in the studied multinational organization, which formed the core of the proposed framework and answered the first research sub-question of what needs to be measured. In Cycle 2, a more in-depth analysis was provided for each of the suggested Key Performance Indicators including the techniques for capturing, calculating, and evaluating the information provided by each KPI. In the course of Cycle 2, the second research sub-question was answered, clarifying how the data for each KPI needed to be measured, interpreted, and acted upon. Consequently, after two incremental research cycles, the primary research question was answered describing the practical framework that may be used for monitoring the performance of agile IT projects delivered in an output-based outsourcing context. This framework was evaluated by the professionals within the context of the studied organization and received positive feedback across all four evaluation criteria set forth in this research, including the low overhead of data collection, high value of provided information, ease of understandability of the metric dashboard, and high generalizability of the proposed framework.

Measuring the performance of corporate events : an empirical study of measuring practices in Finnish companies

Corporate events as an effective part of marketing communications strategy seem to be underestimated in Finnish companies. In the rest of the Europe and the USA, investments in events are increasing, and their share of the marketing budget is significant. The growth of the industry may be explained by the numerous advantages and opportunities that events provide for attendees, such as face-to-face marketing, enhancing corporate image, building relationships, increasing sales, and gathering information. In order to maximize these benefits and return on investment, specific measurement strategies are required, yet there seems to exist a lack of understanding of how event performance should be perceived or evaluated. To address this research gap, this research attempts to describe the perceptions of and strategies for evaluating corporate event performance in the Finnish events industry. First, corporate events are discussed in terms of definitions and characteristics, typologies, and their role in marketing communications. Second, different theories on evaluating corporate event performance are presented and analyzed. Third, a conceptual model is presented based on the literature review, which serves as a basis for the empirical research conducted as an online questionnaire. The empirical findings are to a great extent in line with the existing literature, suggesting that there remains a lack of understanding corporate event performance evaluation, and challenges arise in determining appropriate measurement procedures for it. Setting clear objectives for events is a significant aspect of the evaluation process, since the outcomes of events are usually evaluated against the preset objectives. The respondent companies utilize many of the individual techniques that were recognized in theory, such as calculating the number of sales leads and delegates. However, some of the measurement tools may require further investments and resources, thus restricting their application especially in smaller companies. In addition, there seems to be a lack of knowledge of the most appropriate methods in different contexts, which take into account the characteristics of the organizing party as well as the size and nature of the event. The lack of inhouse expertise enhances the need for third-party service-providers in solving problems of corporate event measurement.

Performance of active portfolio management and the effect of stock market trend

Purpose of the study is to evaluate performance of active portfolio management and the effect of stock market trend on the performance. Theory of efficient markets states that market prices reflect all available information and that all investors share a common view of future price developments. This view gives little room for the success of active management, but the theory has been disputed – at least the level of efficiency. Behavioral finance has developed theories that identify irrational behavior patterns of investors. For example, investment decisions are not made independent of past market developments. These findings give reason to believe that also the performance of active portfolio management may depend on market developments. Performance of 16 Finnish equity funds is evaluated during the period of 2005 to 2011. In addition two sub periods are constructed, a bull market period and a bear market period. The sub periods are created by joining together the two bull market phases and the two bear market phases of the whole period. This allows for the comparison of the two different market states. Performance of the funds is measured with risk-adjusted performance by Modigliani and Modigliani (1997), abnormal return over the CAPM by Jensen (1968), and market timing by Henriksson and Merton (1981). The results suggested that in average the funds are not able to outperform the market portfolio. However, the underperformance was found to be lower than the management fees in average which suggests that portfolio managers are able to do successful investment decisions to some extent. The study revealed substantial dependence on the market trend for all of the measures. The risk-adjusted performance measure suggested that in bear markets active portfolio managers in average are able to beat the market portfolio but not in bull markets. Jensen´s alpha and the market timing model also showed striking differences between the two market states. The results of these two measures were, however, somewhat problematic and reliable conclusions about the performance could not be drawn.