Support vector machines for classifying and describing credit applicants: detecting typical and critical regions

Credit applicants are assigned to good or bad risk classes according to their record of defaulting. Each applicant is described by a high-dimensional input vector of situational characteristics and by an associated class label. A statistical model, which maps the inputs to the labels, can decide whether a new credit applicant should be accepted or rejected, by predicting the class label given the new inputs. Support vector machines (SVM) from statistical learning theory can build such models from the data, requiring extremely weak prior assumptions about the model structure. Furthermore, SVM divide a set of labelled credit applicants into subsets of ‘typical’ and ‘critical’ patterns. The correct class label of a typical pattern is usually very easy to predict, even with linear classification methods. Such patterns do not contain much information about the classification boundary. The critical patterns (the support vectors) contain the less trivial training examples. For instance, linear discriminant analysis with prior training subset selection via SVM also leads to improved generalization. Using non-linear SVM, more ‘surprising’ critical regions may be detected, but owing to the relative sparseness of the data, this potential seems to be limited in credit scoring practice.     

Process and content: two ways of using SSM

Soft systems methodology (SSM) includes several ways of gaining a rich appreciation of the problem situation addressed. ‘Analysis One’, exploration of the intervention itself, is the subject here, since it is sparsely covered in the literature. The analysis is conducted in terms of three roles: ‘client’, ‘problem solver’ and ‘problem owner’. Whoever is in the role of ‘problem solver’ is free to define a list of possible ‘problem owners’, which brings many perspectives to bear on the situation. It was realized that ‘client’ and ‘problem solver’ should themselves feature in the ‘problem owner’ list. The ‘problem’ owned by the ‘problem solver’ is that of undertaking the intervention. This led to a realization that SSM is relevant to both the content of a perceived situation (SSMc) and the process of dealing with that content (SSMp). This development is described and illustrated by work in the National Health Service. The focus of the SSM use was to define the intellectual process for a service specification project which NHS professionals would themselves carry out.     

Developing a Visual Interactive Model for Corporate Cash Management

The operational research/management science journals contain an extensive literature that addresses the corporate cash management problem; yet few, if any, companies make use of any of this published work in their daily cash-management decision making. A review of the literature suggests that the reason for this lack of applications may well be poor problem formulation—the problems that are solved in the literature as ‘cash management’ problems evolve from a ‘hard systems’ view of real-world cash management. However, the problem as perceived by cash managers involves both dynamic and loosely structured components which are difficult to model using classical (i.e. ‘hard systems’) approaches.We therefore decided to approach the cash management problem as an experiment in the use of a novel visual interactive problem solving (VIPS) methodology. The aim of the experiment was to develop an implementable, visual interactive model to support daily cash management decision making. Working closely with a corporate cash manager, we first developed a visual model of his daily decision problem and then agreed on the feasible options and the interactive requirements. At this stage, the problem was sufficiently well defined for a mathematical model to be built and the visual model made ‘smart’.This paper discusses the results of this experiment and suggests that VIPS may have distinct advantages as a problem-solving technique in loosely structured, ‘messy’ problem situations.     

The directed polymer in a random environment

A continuous space/time approximation of the well known ‘directed polymer’ problem is considered. Connection between the ‘Helmholtz Free Energy’ and the ‘Two Walker problem’ is shown. Rigorous proof of the superdiffusive mean squared displacement exponent of 4/3 is given when there is one space dimension and one time dimension. Asymptotically diffusive behaviour of c(k)tis shown when there are one ‘time’ and two ‘space’ dimensions. For higher dimensions, the behaviour is diffusive and the mean squared displacement is asymptotically t d. These results hold for all temperature, because the phase transition in the discrete model is no longer present in the continuous model; the renormalization procedure has set the transition temperature to k _crit =0The joint distribution is also shown to be asymptotically sub-Gaussian for all dimensions and all temperatures (in the sense that the p ^thmoments as a function of pincrease more slowly than the moments of a Gaussian distribution). The ‘Helmholtz Free Energy’ is also calculated for this model and the quenched and annealed free energies are shown to be identical for all temperature     

Budget and long-range planning in public transportation — An applicative linear programming model

Economic and financial planning is an actual problem for Italian Transport Authorities, since in Italy there are scarce financial resources, to cover either the difference between costs and fares proceeds or investment needs. The proposed model is ‘just tailored’ on public transport, subjected to the Italian laws; it takes into account the particular activities (‘functions’) connected with public transport operating (e.g.: fare collection, service production, maintenance, purchases, inventories, administration). Each function is considered as a ‘module’ with its own constraints, and it is connected with the other ‘modules’, so that we can get a ‘representative’ model. Also the most significant parameters of the ‘public transport operating problem’ (e.g.: vehicle miles, passenger journeys, number of vehicles, number of employees, and so on) are connected one another. Since all relations and constraints, connecting the ‘problem variables’ can be represented as linear, the model structure is based on linear programming; this fact allows to pursue an optimum for one or more objective functions, each of them identifing an operational policy. So it is possible to connect simulation with optimization. A multi-period model can be used for long-range planning, pursuing a multi-year optimum; this is the most significant use. The Administrative Module contains Balance-Sheet, Cash-Flow, Profit and Loss Account, which are ‘constraints’ in the model and are expressed as Italian financial laws require; this structure is one of the main characteristics. Modules' dimensions and additional constraints (e.g. economic or financial ratios, fare level, turnover of inventories, and so on) can be ‘tailored’ on each particular case, driving the model towards more realistic solutions.     

Multi-Skilled Workforce Optimisation

This paper describes a problem faced by CS Energy's Swanbank Power Station in the Australian state of Queensland. It involved the personnel scheduling (rostering) of staff with multiple skill levels at the power station. Such a problem can be classified using the six stage construction process proposed by Ernst et al. We assume that the three processes of ‘demand modelling,’ ‘shift starting times’ and ‘task scheduling’ are specified. We are concerned with the essential processes of ‘day off scheduling,’ ‘line of work construction’ and ‘shift assignment to staff’ with requirements to maintain multiple skills. Several other authors have reported results for staff with hierarchical skills while the methods proposed in this paper are for non-hierarchical skill sets. The paper describes a set covering approach to the multi-skilled rostering problem. We propose a number of solution strategies for the set covering approach and give a comparison of the results.     

Soft with a hard centre: discrete-event simulation in facilitation

Discrete event simulation is normally described as a ‘hard’ OR technique. This may not, however, always be the case. An example of a simulation of a user support helpline is described which, it is argued, has many of the traits of a ‘soft’ OR intervention. In particular, the study involved a facilitated discussion around a simulation model about possible improvements to a problem situation. The nature of the intervention is considered from both a methodological and paradigmatic perspective, and conclusions are drawn about where the intervention lies on a ‘hard’ to ‘soft’ continuum. It is argued that ‘soft’ issues need to be subsumed into the prescribed methodology for discrete-event simulation.     

Selecting the best choice in the weighted secretary problem

In the sequential evaluation and selection problem with n applicants, we assume that a decision maker has some prior information about each applicant so that unequal weights may be assigned to each applicant according to his or her likelihood of being the best among all applicants. Assuming that the pre-assigned weights are available in advance, we derive the optimal selection strategy that maximizes the probability of selecting the best among all applicants. For the case where the decision maker is permitted to rearrange the sequence in which applicants are evaluated, we further propose a simple heuristic procedure to the problem of optimally ordering the sequence of evaluations. Based on a pairwise comparison matrix and a goal programming procedure, we also propose a method that easily computes the weights in a practical situation.     

Dense sets and the projection theorem for acoustic harmonic waves in a homogeneous finite depth ocean

The scattering of a ‘plane wave’ off a submerged body situated in an ocean of finite depth is investigated. The index of refraction is considered to be depth-independent. It is shown that the far field is not unique; hence, the problem of determining the shape of an object from its far field is not well-posed. If solutions are sought among a restricted class of problems the ‘dense set’ property implies that the problem can be made well-posed.     

Formal optimization of some reduced linear programming problems

It is often possible (and profitable) to reduce or ‘Presolve’ linear programs. In particular, there are frequently constraints which force many of the variables to be at bound. Unfortunately, the solution found by the simplex method for such reduced models is not usually ‘formally’ optimal, in the sense that nonoptimal dual values may be present when the original problem is restored. Furthermore, the restored (full) problem is now totally degenerate, and may require many iterations to achieved formal optimality. We describe an efficient ‘Postsolve’ procedure for attaining the formal optimum solution, and give computational results.     

‘Minesweeper’ and spectrum of discrete Laplacians

The article is devoted to a problem inspired by the ‘Minesweeper’ computer game. It is shown that certain configurations of open cells guarantee the existence and the uniqueness of solution. Mathematically the problem is reduced to some spectral properties of discrete differential operators. It is shown how the uniqueness can be used to create a new game which preserves the spirit of ‘Minesweeper’ but does not require a computer.     

Probability Modelling and Optimal Location of a Travelling Salesman

The ‘random part’ of an operations research model may be less satisfactory than the ‘deterministic part’, and it may thus be desirable to design algorithms that require few probability assumptions and make few calls to a suitable ‘probability oracle’. We consider here the problem of locating a service facility on a tree network so as to minimize the expected length of a travelling salesman tour through a random set of demand nodes.     

Solving 0-1 Minimax Problems

It has been shown that it is possible to solve a minimax 0-1 programming problem by transforming the objective function coefficients and solving a minisum problem. This result is very useful for solving a ‘hard’ minimax problem when there exists a ‘relatively easy’ algorithm to solve the minisum problem. We discuss such problems and present a new transformation.     

Sequential fuzzy system identification

The problem of deriving the structure of a non-deterministic system from its behaviour is a difficult one even when that behaviour is itself well-defined. When the behaviour can be described only in fuzzy terms structural inference may appear virtually impossible. However, a rigorous formulation and solution of the problem for stochastic automata has recently been given [1] and, in this paper, the results are extended to fuzzy stochastic automata and grammars. The results obtained are of interest on a number of counts. (1) They are a further step towards an integrated ‘theory of uncertainty’; (2) They give new insights into problems of inductive reasoning and processes of ‘precisiation’; (3) They are algorithmic and have been embodied in a computer program that can be applied to the modelling of sequential fuzzy data; (4) They demonstrate that sequential fuzzy data may be modelled naturally in terms of ‘possibility’ vectors.     

Exceptional Families of Elements for Optimization Problems in Reflexive Banach Spaces with Applications

In this paper, we propose a new notion of ‘exceptional family of elements’ for convex optimization problems. By employing the notion of ‘exceptional family of elements’, we establish some existence results for convex optimization problem in reflexive Banach spaces. We show that the nonexistence of an exceptional family of elements is a sufficient and necessary condition for the solvability of the optimization problem. Furthermore, we establish several equivalent conditions for the solvability of convex optimization problems. As applications, the notion of ‘exceptional family of elements’ for convex optimization problems is applied to the constrained optimization problem and convex quadratic programming problem and some existence results for solutions of these problems are obtained.     

On efficient stopping times

This paper is devoted to efficient sequential estimation in stochastic processes whose corresponding sufficient statistics are processes with stationary independent increments. It is proved that a stopping time is efficient if and only if it represents a time of the first attaining of a hyperplane., which cannot ‘be passed’, in the sense which is made precise below. The problem of determining the explicit form of the hyperplanes which cannot ‘be passed’ is also discussed.     

On possibilistic linear programming

The problem of processing and combining possibilistic information in the scope of linear programming is approached. As apparatus to cope with this problem, we use the concepts of ‘more possible value’, ‘α-possibly feasible action’ and ‘α-possibly efficient action’.Issues regarding the feasibility of the resulting deterministic programs as well as the characterization of their optimal solutions are also discussed.     

A useful transform of standard input data for a classical NP-complete problem

The archetypal symmetric travelling salesman problem can be seen in a new and interesting way, by using first a standard preparatory phase of input data, and then by applying a transform from the set D of ‘distances’ among ‘cities’ and the set B of ‘loss of optimality’.The specific form of D→B transform is introduced and discussed. In order to show in realistic terms the interest of the approach proposed, a class of ‘diffusive’ heuristic procedures operating from B is defined.An example of solution by an algorithm included in this class is completely worked out; an outline of computational tests done on the same algorithm is also given.     

The fuzzy decision problem: An approach to the point estimation problem with fuzzy information

This paper is devoted to the extension of the Bayesian method for the point estimation, when the available information is ‘vague’.In the nonfuzzy case, the parametric estimation can be approached as a particularization in the statistical decision problem. This motivates us to accomplish the mentioned extension by looking at the parametric estimation in the fuzzy case as a special situation in the fuzzy decision problem (defined by Tanaka, Okuda and Asia).In this way, concepts in the fuzzy decision problem are first ‘expressed’ in the estimation terminology. Then, on the basis of these concepts, we shall introduce some notions and state some interesting results. Finally, several illustrative examples will be exposed.