Passenger%2Dbased predictive modeling of airline no%2Dshow rates.
Airlines routinely overbook flights based on the expectation that some fraction of booked passengers will not show for each flight. Accurate forecasts of the expected number of noshows for each flight can increase airline revenue by reducing the number of spoiled seats (empty seats that might otherwise have been sold) and the number of involuntary denied boardings at the departure gate. Conventional no-show forecasting methods typically average the no-show rates of historically similar flights, without the use of passenger-specific information.
Data%2Dintensive analytics for predictive modeling.
The Data Abstraction Research Group was formed in the early 1990s, to bring focus to the Mathematical Sciences department’s work in the emerging area of Knowledge Discovery and Data Mining (KD&amp;DM). Most activities in this group have been performed in the technical area of predictive modeling, roughly at the intersection of machine learning, statistical modeling, and database technology. There has been a major emphasis on using business and industrial problems to drive the research agenda. Major accomplishments include advances in methods for feature analysis, rule based pattern discovery and probabilistic modeling, and novel solutions for insurance risk management, targeted marketing, and text mining. This paper presents an overview of the group’s major technical accomplishments. 1
Ensemble Modeling Through Multiplicative Adjustment of Class Probability.

Preface.
part of a series of studies in support of the California Governor’s Advisory Committee on Education Excellence. i PREFACE i LIST OF FIGURES iv
A New Approach for Item Choice Recommendations.

Advances in predictive models for data mining.
Expanding application demand for data mining of massivedatawarehouses has fueled advances in automated predictive methods. We examine a few successful application areas and their technical challenges. We review the key theoretical developments in PAC and statistical learning theory that have lead to the development of support vector machines and to the use of multiple models for increased predictive accuracy.  Keywords: data mining, text mining, machine learning, boosting  1 Introduction  Predictive modeling, which is perhaps the most-used subfield of data mining, draws from statistics, machine learning, database techniques, pattern recognition, and optimization techniques. The proven utility of industrial applications has led to advances in predictive modeling. We will review a few application areas that have demonstrated the importance of predictive modeling and have also fueled advances. Some importantadvances in learning theory will be considered in section 3. The idea of using mul...
AI at IBM Research.
IBM has played an active role in AI research since the field&#039;s inception more than 50 years ago. In a trend that reflects the increasing demand for applications that behave intelligently, IBM today carries out most AI research in an interdisciplinary fashion by combining AI techniques with other computing techniques to solve difficult technical problems.  1 
Advanced in Predictive Data Mining Methods.

Partitioning Nominal Attributes in Decision Trees.
. To find the optimal branching of a nominal attribute at a node in an  L-ary decision tree, one is often forced to searchover all possible L-ary partitions for the one that yields the minimum impurity measure. For binary trees (L =2) when there are just two classes a short-cut search is possible that is linear in n,the  number of distinct values of the attribute. For the general case in which the number of classes, k,may be greater than two, Burshtein et al. haveshown that the optimal partition satisfies a condition that involves the existence of  i  L  2  j  hyperplanes in the class probability space. We derive a property of the optimal partition for concave impurity measures (including in particular the Gini and entropy impurity measures) in terms of the existence of L vectors in the dual of the class probability space, which implies the earlier condition. Unfortunately, these insights still do not offer a practical search method when n  and k are large, even for binary trees. We th...
Decomposition of Heterogeneous Classification Problems.
In some classification problems the feature space is heterogeneous in that the best features on which to base the classification are different in different parts of the feature space. In some other problems the classes can be divided into subsets such that distinguishing one subset of classes from another and classifying examples within the subsets require very different decision rules, involving different sets of features. In such heterogeneous problems, many modeling techniques (including decision trees, rules, and neural networks) evaluate the performance of alternative decision rules by averaging over the entire problem space, and are prone to generating a model that is suboptimal in any of the regions or subproblems. Better overall models can be obtained by splitting the problem appropriately and modeling each subproblem separately. This paper presents a new measure to determine the degree of dissimilarity between the decision surfaces of two given problems, and suggests a way to search for a strategic splitting of the feature space that identifies regions with different characteristics. We illustrate the concept using a multiplexor problem, and apply the method to a DNA classification problem. 
Decomposition of Heterogeneous Classification Problems.
In some classification problems the feature space is heterogeneous in that the best features on which to base the classification are different in different parts of the feature space. In some other problems the classes can be divided into subsets such that distinguishing one subset of classes from another and classifying examples within the subsets require very different decision rules, involving different sets of features. In such heterogeneous problems, many modeling techniques (including decision trees, rules, and neural networks) evaluate the performance of alternative decision rules by averaging over the entire problem space, and are prone to generating a model that is suboptimal in any of the regions or subproblems. Better overall models can be obtained by splitting the problem appropriately and modeling each subproblem separately. This paper presents a new measure to determine the degree of dissimilarity between the decision surfaces of two given problems, and suggests a way to search for a strategic splitting of the feature space that identifies regions with different characteristics. We illustrate the concept using a multiplexor problem, and apply the method to a DNA classification problem. 
R%2DMINI An Iterative Approach for Generating Minimal Rules from Examples.
Generating classification rules or decision trees from examples has been a subject of intense study in the pattern recognition community, the statistics community and the machine learning community of the artificial intelligence area. We pursue a point of view that minimality of rules is important, perhaps above all other considerations (biases) that come into play in generating rules. We present a new minimal rule generation algorithm called R-MINI (Rule-MINI) that is an adaptation of a well-established heuristic switching function minimization technique, MINI. The main mechanism that reduces the number of rules is repeated application of generalization and specialization operations to the rule set while maintaining completeness and consistency. R-MINI results on some benchmark cases are also presented.  I. Introduction  There are many approaches to generating Disjunctive Normal Form (DNF) rules from examples. The Aq family of rule generation and other approaches [1-4] incrementally c...
Use of Contextaul Information for Feature Ranking and Discretization.

Predicting Equity Returns from Securities Data.
Our experiments with capital markets data suggest that the domain can be effectively modeled by classification rules induced from available historical data for the purpose of making gainful predictions for equityinvestments. New classification techniques developed at IBM Research, including minimal rule generation (R-MINI) and contextual feature analysis, seem robust enough for consistently extracting useful information from noisy domains such as financial markets. We will briefly introduce the rationale for our minimal rule generation technique, and the motivation for the use of contextual information in analyzing features. We will then describe our experience from several experiments with the S&amp;P 500 data, illustrating the general methodology, and the results of correlations and simulated managed investment based on classification rules generated by R-MINI. Wewillsketchhow the rules for classifications can be effectively used for numerical prediction, and eventually to an investment ...
Predicting Equity Returns from Securities Data with Minimal Rule Generation.
Our experiments with capital markets data suggest that the domain can be effectively modeled by classification rules induced from available historical data for the purpose of making gainful predictions for equityinvestments. New classification techniques developed at IBM Research, including minimal rule generation (R-MINI) and contextual feature analysis, seem robust enough for consistently extracting useful information from noisy domains such as financial markets. We will briefly introduce the rationale for our minimal rule generation technique, and the motivation for the use of contextual information in analyzing features. We will then describe our experience from several experiments with the S&amp;P 500 data, illustrating the general methodology, and the results of correlations and simulated managed investment based on classification rules generated by R-MINI. Wewillsketchhow the rules for classifications can be effectively used for numerical prediction, and eventually to an investment ...
Guest Editors%27 Introduction Enabling Technology for Knowledge%2DBased Systems Development.

Using Qualitative Reasoning to Understand Financial Arithmetic.

A Knowledge%2DBased Consultant for Financial Marketing.

Automation of MVS Operations %2D an Expert%2DSystems Approach.

A Continuous Real%2DTime Expert System for Computer Operations.

Expert Systems in Engineering %2D Guest Editor%27s Introduction.

YES%2FMVS A Continuous Real Time Expert System.

Automation of MVS Operations%2C an Expert Systems Approach.

A Simple Procedure to Generate Optimum Test Patterns for Parity Logic Networks.

Fault Analysis and Test Generation for Programmable Logic Arrays %28PLA%27s%29.

An Algorithm for Transformation of an Arbitrary Switching Function to a Completely Symmetric Function.

Codes for Self%2DClocking
Abstract: We consider NRZI waveform codes that satisfy a given set of run-length constraints and the upper bound on the accumulated dc charge of the waveform. These constraints enable the codeword to be self-clocking, ac-coupled, and suitable for data processing tape and communication applications. Various aspects of synthesis and analysis of such codes, called (d, k, C) codes, are illustrated by means of several examples. The choice of the initial state of the encoder is shown to influence the length of the data sequence over which the encoder must look-ahead.

MINI A Heuristic Approach for Logic Minimization.
Abstract: MINI is a heuristic logic minimization technique for many-variable problems. It accepts as input a Boolean logic specification expressed as an input-output table, thus avoiding a long list of minterms. It seeks a minimal implicant solution, without generating all prime implicants, which can be converted to prime implicants if desired. New and effective subprocesses, such as expanding, reshaping, and removing redundancy from cubes, are iterated until there is no further reduction in the solution. The process is general in that it can minimize both conventional logic and logic functions of multi-valued variables.



