Miles Kennedy

An Application of Rule-Based Forecasting to a Situation Lacking Domain Knowledge

Social Science Research Network, Feb 7, 2005

Rule-based forecasting (RBF) uses rules to combine forecasts from simple extrapolation methods. W... more Rule-based forecasting (RBF) uses rules to combine forecasts from simple extrapolation methods. Weights for combining the rules use statistical and domain-based features of time series. RBF was originally developed, tested, and validated only on annual data. For the M3-Competition, three major modifications were made to RBF. First, due to the absence of much in the way of domain knowledge, we prepared the forecasts under the assumption that no domain knowledge was available. This removes what we believe is one of RBF's primary advantages. We had to re-calibrate some of the rules relating to causal forces to allow for this lack of domain knowledge. Second, automatic identification procedures were used for six time-series features that had previously been identified using judgment. This was done to reduce cost and improve reliability. Third, we simplified the rule-base by removing one method from the four that were used in the original implementation. Although this resulted in some loss in accuracy, it reduced the number of rules in the rule-base from 99 to 64. This version of RBF still benefits from the use of prior findings on extrapolation, so we expected that it would be substantially more accurate than the random walk and somewhat more accurate than equal weights combining. Because most of the previous work on RBF was done using annual data, we especially expected it to perform well with annual data.

Download

Automatic Identification of Time Series Features for Rule-Based Forecasting

Social Science Research Network, Feb 7, 2005

Rule-based forecasting (RBF) is an expert system that uses features of time series to select and ... more Rule-based forecasting (RBF) is an expert system that uses features of time series to select and weight extrapolation techniques. Thus, it is dependent upon the identification of features of the time series. Judgmental coding of these features is expensive and the reliability of the ratings is modest. We developed and automated heuristics to detect six features that had previously been judgmentally identified in RBF: outliers, level shifts, change in basic trend, unstable recent trend, unusual last observation, and functional form. These heuristics rely on simple statistics such as first differences and regression estimates. In general, there was agreement between automated and judgmental codings for all features other than functional form. Heuristic coding was more sensitive than judgment and consequently, identified more series with a certain feature than judgmental coding. We compared forecast accuracy using automated codings with that using judgmental codings across 122 series. Forecasts were produced for six horizons, resulting in a total of 732 forecasts. Accuracy for 30% of the 122 annual time series was similar to that reported for RBF. For the remaining series, there were as many that did better with automated feature detection as there were that did worse. In other words, the use of automated feature detection heuristics reduced the costs of using RBF without negatively affecting forecast accuracy.

Download

The Problem of Neural Networks in Business Forecasting:An Attempt to Reproduce the Hill, Oâ€™Connor and Remus Study

The results from applying neural networks to business forecasting have been mixed. Among the most... more The results from applying neural networks to business forecasting have been mixed. Among the most encouraging efforts is that of Hill, Oâ Connor and Remus (1996). In that study, neural networks produced forecasts that were significantly better than those produced by traditional methods for quarterly and monthly series, and no worse for annual series. We have attempted to reproduce that study. The pattern of our results matches that of the original study and supports many of its conclusions, but we were not able to obtain nearly the magnitude of the improvements reported there. We observe that the stopping rules proposed in the original study have exceedingly high variance associated with them, and do not seem to be reasonable stopping rules for such applications. We conclude that the original studyâ s networks, and in particular its stopping rules, are not described in a way that permits reproducing them. This undermines the value of these systems in practice, since it is difficult both to reproduce them and to identify when and why they are failing to produce results like those obtained in research settings. At the same time, Hill et al.â s conclusion that neural networks represent a good approach to extrapolating nonlinear and discontinuous series is supported.

Download

Self-powered illumination device

Portable arch building structure

Interactive Computer Modeling, Monitoring, and Control of Melbourne's Water Supply System

Water Resources Research, 1985

The Melbourne and Metropolitan Board of Works in Melbourne, Australia operates two principle mode... more The Melbourne and Metropolitan Board of Works in Melbourne, Australia operates two principle models of its water supply system. The first of these is a headworks model that simulates the major reservoirs and transfer mains of Melbourne's water supply system. This model uses monthly data for both the long‐term planning of the water supply system and in assisting in operational decisions regarding transfers of water between headworks storages. The second model is a network model that solves for heads and flows within Melbourne's water supply distribution system. The model can be run over a 24‐hour period and is used for investigating ways of improving the distribution of water within the system and for assisting and training operational staff. This paper describes the application of this network model to Melbourne's water supply system, including the implementation of computer graphics techniques. Examples demonstrating the application of color graphics to different parts ...

Download

Centennial History of The Pennsylvania Railroad Company. By George H. Burgess and Miles C. Kennedy. Foreword by Martin W. Clement. Philadelphia: The Pennsylvania Railroad Company, 1949. Pp. xxvi, 835. $3.50

The Journal of Economic History, 1950

Despite its self-imposed limits, this massive volume is a landmark in the literature of railway h... more Despite its self-imposed limits, this massive volume is a landmark in the literature of railway history. In my opinion it is the most comprehensive, informative, and accurate description of a single system yet to appear. As clearly stated in Martin Clement's Foreword and in the authors' Preface, this centennial history is a factual account of the origins, expansion, operation, and administration of the Pennsylvania. Interpretive conclusions are rare, and except by implication no attempt has been made to integrate the story of the railroad with that of the territory or the nation it has served or to compare it with other railroads or the industry as a whole. Since this is an "official" history, based almost exclusively on, the Pennsylvania's corporate archives and published by the railroad, it represents the company's own analysis of its achievements and does not pretend to do otherwise. The complex story is meticulously organized in chronological fashion. Like the railroad, the book is a heavy, serious, and businesslike affair. In appraising any volume it is essential to judge it first and foremost in the light of its stated purpose. According to this criterion, Messrs. Burgess and Kennedy have done, on the whole, a magnificent job. From a quantitative standpoint, miraculously enough, they have accounted for each foot of track and every corporate unit, every major improvement, and all principal security issues. They have set out in considerable detail the corporation's key policies concerning construction and expansion, operating organization, financing, and competitors and have said something, though not nearly so much, about the company's position over the years toward labor, government, patrons, and the public. One lengthy well-illustrated appendix describes rolling stock, and others give precise information about personnel and financial results. The book is amply provided with clear maps and instructive prints and photographs. There is no bibliography worthy of the name, but it is abundantly apparent that the authors, as they say in their Preface, have had unlimited access to the official minute books and other source records of all the constituent companies; documentation, however, is limited to occasional (and incomplete) references to the few outside sources consulted. The Index, in content and format, is excellent. From a qualitative standpoint the authors have emphasized the facts and trends that they-and presumably the railroad-feel have been the most significant over the long run. With much of their allocation of attention there can be no reasonable disagreement. For example, exactly 40 per cent of the book is devoted to the administration of J. Edgar Thomson, who served as President from 1852 to 1874; in this particular case the reason for such extended treatment is supplied by former President Clement in his Foreword: "More than any other man," he explains, "he [Thomson] shaped the principles and defined the aims and objectives which have been responsible for the Pennsylvania Railroad system as it exists today.. .. With few exceptions, the territory covered. .. at the close of his administration. .. is the same as now, and the development, improvement and refinement of the System since 1874 n a v c

Download

Dynamics for Establishing the Flows of Information Needed for Effective Planning, Administration and Control

Working paper (Sloan School of …, 1974

Page 1. Page 2. LIBRARY OF THE MASSACHUSETTS INSTITUTE OF TECHNOLOGY Page 3. Page 4. Page 5. Page... more

Bruce, Alfred W. The Steam Locomotive in America: Its Development in the Twentieth Century. New York: W. W. Norton & Company, 1952

2000b), “Automatic identification of time series features for rule-based forecasting,” Working paper available from Monica Adya (adya@umbc.edu

Rule-based forecasting (RBF) is an expert system that uses features of time series to select and ... more Rule-based forecasting (RBF) is an expert system that uses features of time series to select and weight extrapolation techniques. Thus, it is dependent upon the identification of features of the time series. Judgmental coding of these features is expensive and the reliability of the ratings is modest. We developed and automated heuristics to detect six features that had previously been judgmentally identified in RBF: outliers, level shifts, change in basic trend, unstable recent trend, unusual last observation, and functional form. These heuristics rely on simple statistics such as first differences and regression estimates. In general, there was agreement between automated and judgmental codings for all features other than functional form. Heuristic coding was more sensitive than judgment and consequently, identified more series with a certain feature than judgmental coding. We compared forecast accuracy using automated codings with that using judgmental codings across 122 series. ...

Download

This journal ar

Rule-based forecasting (RBF) uses rules to combine forecasts from simple extrapolation methods. W... more Rule-based forecasting (RBF) uses rules to combine forecasts from simple extrapolation methods. Weights for combining the rules use statistical and domain-based features of time series. RBF was originally developed, tested, and validated only on annual data. For the M3-Competition, three major modifications were made to RBF. First, due to the absence of much in the way of domain knowledge, we prepared the forecasts under the assumption that no domain knowledge was available. This removes what we believe is one of RBF's primary advantages. We had to re-calibrate some of the rules relating to causal forces to allow for this lack of domain knowledge. Second, automatic identification procedures were used for six time-series features that had previously been identified using judgment. This was done to reduce cost and improve reliability. Third, we simplified the rule-base by removing one method from the four that were used in the original implementation. Although this resulted in so...

Case Western Reserve University

Rule-based forecasting (RBF) is an expert system that uses features of time series to select and ... more Rule-based forecasting (RBF) is an expert system that uses features of time series to select and weight extrapolation techniques. Thus, it is dependent upon the identification of features of the time series. Judgmental coding of these features is expensive and the reliability of the ratings is modest. We developed and automated heuristics to detect six features that had previously been judgmentally identified in RBF: outliers, level shifts, change in basic trend, unstable recent trend, unusual last observation, and functional form. These heuristics rely on simple statistics such as first differences and regression estimates. In general, there was agreement between automated and judgmental codings for all features other than functional form. Heuristic coding was more sensitive than judgment and consequently, identified more series with a certain feature than judgmental coding. We compared forecast accuracy using automated codings with that using judgmental codings across 122 series. ...

Download

Composant de construction de panneau et construction d'un abri

L'invention concerne une structure de construction possedant une structure de charpente tubul... more L'invention concerne une structure de construction possedant une structure de charpente tubulaire en forme de hutte pouvant avoir un parement sous la forme de panneaux courbes pour la realisation d'un abri provisoire ou de secours. Ces panneaux peuvent etre constitues de deux feuilles plastiques espacees reliees ensemble par des nervures paralleles tenant lieu d'armatures pour la realisation de cannelures paralleles dans celles-ci, par exemple, un systeme de feuilles de type Coroplast™. Un tel systeme de feuilles est de preference rendu rigide par une ou plusieurs planches qui sont inserees au travers des cannelures creuses du panneau, ce qui permet de les utiliser en tant que parement semi-rigide sur l'integralite de la structure de construction. Les organes tubulaires exterieurs du châssis sont tous courbes pour ameliorer la rigidite. Le châssis comprend des cannelures pour asseoir les panneaux ainsi que contribuer a l'integrite de structure de l'abri.

Engineering, education and energy: Some contributions of Alan Lee Titchener

A distinguished member of IPENZ, Emeritus Professor A L Titchener, died on 23 June 1994. Alan Tit... more A distinguished member of IPENZ, Emeritus Professor A L Titchener, died on 23 June 1994. Alan Titchener contributed many papers to the Institution's journals, and received many prizes and awards - most notably, the F W MacLean Citation in 1986. Professor Titchener retired from the School of Engineering at the University of Auckland at the beginning of 1986, having spent 34 years there as Lecturer, Senior Lecturer, Associate Professor, then Professor and Head of the Department of Mechanical Engineering; Foundation Professor and Head of the Department of Chemical and Materials Engineering; and Dean of Engineering on two separate occasions. The School arranged a seminar on 6 September 1985 to honour Alan Titchener and two colleagues, Arved Raudkivi and Peter Taylor. The following contribution is based on a presentation at that seminar, and touches on aspects of Alan Titchener's work which lay outside the mainstream of engineering.

Panel Building Component and Building Shelter

Portable arch building structure

The Politics of Space: Poetical Dwelling and the Occupation of Poetry

Spatiality and Symbolic Expression, 2015

In this chapter, by attempting to work within a specific version of the general framework of four... more In this chapter, by attempting to work within a specific version of the general framework of four zones outlined in the Introduction to this volume, I critique what I term the Heideggerian conception of Holderlin’s poetical dwelling, in order to examine and question the effaced underpinnings of much of what has been said about spatiocultural studies up to this point. An examination of Tom Paulin’s poetic riposte to this Heideggerian occupation as articulated in his 2002 poetry collection, The Invasion Handbook, is taken as an act of counter-occupation that points toward a more authentic and politically committed appraisal of poetical dwelling and human occupation itself.

Self-powered illumination device

A protocol study of a complexity model of information acquisition from graphs and tables

A growing body of research on the use of computer graphics for decision making has failed to shed... more A growing body of research on the use of computer graphics for decision making has failed to shed much light on the issue of the superiority of graphics over other forms of presentation or of one format over another. The general consensus now is that the effectiveness of a particular format is primarily determined by its appropriateness to the particular task. The result is a growing body of heuristics which lack a unifying theory. In most studies the process of information acquisition from graphs is treated as a black box. This dissertation studies the question of the task motivated processes used to acquire information from data displays including graphs and tables. A model of these processes is developed which synthesizes a number of approaches from various reference disciplines including human information processing and cognitive science. The model argues that each information acquisition task comprises a predictable series of subtasks or elementary information processes. Each subtask involves the perceptual processing of a number of information cues. The task, format, complexity of the data and requisite cognitive effort all determine the number and complexity of these subtasks. The most efficient format is the one which minimizes the aggregate cognitive complexity as measured by the sum of the efforts required for each of the constituent subtasks. The model leads to improved user interfaces where the user specifies queries in terms of the task and the system selects the most efficient form of display. A protocol study of the research model was conducted using a novel approach which overcomes many of the problems associated with these studies. Subjects were presented with a series of displays and questions and asked to trace the area of interest with a pen. Regression analysis was used to develop coefficients for each subtask equivalent to the average time required to perform that subtask. The resulting normative model can be used to predict relative performance times for a variety of formats for a given task, subject and data set. The signs and magnitudes of the estimated coefficients generally conformed to expectations. In a second experiment, the model performed well as a predictor of relative times.

The Politics of Space

Spatiality and Symbolic Expression

Uploads

Papers by Miles Kennedy

Log In