UNIVERSITI PUTRA MALAYSIA FORECASTING THE GROWTH OF MANUFACTURING INDUSTRY IN MALAYSIA USING ARTIFICIAL NEURAL NETWORK (ANN)

Transcription

1 UNIVERSITI PUTRA MALAYSIA FORECASTING THE GROWTH OF MANUFACTURING INDUSTRY IN MALAYSIA USING ARTIFICIAL NEURAL NETWORK (ANN) NORZAINI ZAKARIA. FSKTM

2 FORECASTING THE GROWTH OF MANUFACTURING INDUSTRY IN MALAYSIA USING ARTIFICIAL NEURAL NETWORK (ANN) NORZAINI BINTI ZAKARIA Thesis submitted in Partial Fulfillment of Requirement for the Degree of Master of Science in the Faculty of Computer Science and Information Technology Universiti Putra Malaysia December 2006

3 ABSTRACT Abstract of thesis presented to the Senate of Universiti Putra Malaysia in fulfillment of the requirements for the Degree of Master of Science. FORECASTING THE GROWTH OF MANUFACTURING INDUSTRY IN MALAYSIA USING ARTIFICIAL NEURAL NETWORK (ANN) NORZAINI BINTI ZAKARIA December 2006 An Artificial Neural Network (ANN) formulation for forecasting the growth of manufacturing industries in Malaysia is presented. The effectiveness of ANN for predicting one criterion, Y form several predictors, X, is investigated. Comparative studies are examined between the result of the predictions from the ANN model trained with Multi-Layer Feedforward Perception, a Generalised Regressions Neural Network (GRNN) algorithm and the result obtained from traditional statistical approach.

4 Abstrak ini dibentangkan kepada Senat Universiti Putra Malaysia sebagai memenuhi keperluan untuk Sarjana Sains PERAMALAN INDUSTRI PEMBUATAN DI MALAYSIA DENGAN PERISIAN RANGKAIAN NEURAL Oleh NORMI BINTI ZAKA.RIA Laporan ini membincangkan penggunaan kaedah rangkaian neural bagi menganalisa aliran pertumbuhan industri pembuatan di Malaysia pada masa hadapan. Kajian ini adalah berbentuk regresi pelbagai (Multiple Regression) yang akan meramal pembolehubah bersandar Y dari pelbagai pembolehubah talc bersandar Xi. Projek ini akan meramal nilai ditambah bagi 2 tahun perlunjuran iaitu tahun 2004 dan 2005 dan akan membandingkan hasil dari kaedah "Multilayer Feedforward Perception" dan "Generalised Regression" dalam Neural Network dengan kaedah statistik. iii

5 ACKNOWLEDGMENT All praise to Allah S.W.T, who has showered me with patience and bless to finish my Master thesis. Alhamdulillah. I would like to express my sincere gratitude to Dr. Norwati binti Mustapha for being the guiding force behind the research necessary for the completion of this thesis. Without her support, guidance, encouragement and idea this work would not have been possible. I would like to acknowledge the support of my colleagues at Universiti Kuala Lumpur especially Pn. Yasmin Yahya for the help and support towards the completion of my thesis. I am enormously grateful to my family and friends for their support and understanding while I completed this thesis. I especially thank my husband, Rosli Kassim, for providing an endless supply of support that never ceased to inspire me. And, finally, I offer thanks to my children, Muhammad Firdaus, Nurul Izzah and Azizah for their attempt to understand my intention to further my study. Norzaini Zakaria Faculty of Computer Science and Information Technology Universiti Putra Malaysia December 2006

6 APPROVAL This thesis submitted to Faculty of Computer Science and Information Technology of Universiti Putra Malaysia and has been accepted as fulfillment of the requirements for the Degree of Master of Computer Science. Dr. Norwati Binti Mustapha Department of Computer Science Faculty of Computer Science and Information Technology Universiti Putra Malaysia Date :

7 DECLARATION I hereby declare that the thesis is based on my original work except for quotations and citations which have been duly acknowledged. I also declare that it has not been previously or concurrently submitted for any other degree at Universiti Putra Malaysia or other institutions. Norzaini Binti Zakaria

8 CONTENTS Page ABSTRACT ABSTRAK ACKNOWLEDGMENT APPROVAL DECLARATION CONTENTS LIST OF FIGURES LIST OF TABLES LIST OF ABBREVIATION ii iii iv v vi vii X xii xiii CHAPTER I INTRODUCTION 1.1 BACKGROUND 1.2 PROBLEM STATEMENT 1.3 OBJECTIVE 1.4 PROJECT SCOPE 1.5 ORGANISATION OF THE STUDY CHAPTER 11 LITERATURE REVIEW 2.1 INTRODUCTION 2.2 MANUFACTURING INDUSTRY IN MALAYSIA 2.3 RELATED WORKS vii

9 2.4 SUMMARY CHAPTER 111 RESEARCH METHODOLOGY INTRODUCTION SYSTEM DEVELOPMENT APPROACH RESEARCH METHODOLOGY - PHASE RESEARCH METHODOLOGY - PHASE 2 INPUT AND OUTPUT SPECIFICATION INPUT OUTPUT SOFTWARE EVIEWS NEURAL NETWORKS WITH NEURALTOOLS 3.5 SUMMARY CHAPTER IV DATA ANALYSIS AND MODEL SPECIFICATION 4.1 INTRODUCTION 4.2 MODEL SPECIFICATION 4.3 STATISTICAL MODEL DECSRIPTIVE STATISTICS PLOTTING THE VARIABLES MODEL FORMUMON MULTICOLLINEARITY TEST viii

10 HETEROSKEDASTICITY TEST AUTOCORRELATION TEST 4.4 NEURAL NETWORK MODEL - MLFP AND GRNN MODEL ARCHtTEEHTURE 4.5 SUMMARY CHAPTER V RESULT AND DISCUSSION 5.1 INTRODUCTION 5.2 EMPERICAL RESULTS RESULT FROM STATISTICAL APPROACH RESULT FROM NEURAL NETWORK APPROACH 3.3 SUMMARY CHAPTER VI CONCLUSION AND FUTERE WORKS 6.1 CONCLUSION 6.2 WTURE WORKS REFERENCES APPENDIX A APPENDIX B

11 LIST OF FIGURES Figure No Description Data Mining Step of Knowledge Discovery (KDD) Example of user interface in EViews Architecture of Neural Network Unit with its weights and bias Sigrnoid activation function Summary statistics of all variables Value Added Employment / Labour Export Investment Value Added vs Time Employment vs Time Export vs Time Investment vs Time Value Added vs Employment Value Added vs Export Value Added vs Investment LVA vs LEMP LVA vs LEKS LVA vs LINV Page

12 Figure No Description Correlation between predicted variables Result of Variance Inflation Factor ~esidual' vs LVA White Heteroskedasticity Test for model 1 White Heteroskedasticity Test for model 2 Residual vs Time Durbin-Watson Test for model 1 Durbin-Watson Test for model 2 Durbin-Watson Test for model 3 Durbin-Watson Test for model 4 Model estimation for each model Results from NeuralTools MLFP Architecture for Model 2 (MLFP) Model estimation using Statistical approach Results from NeuralTools Predicted Value Added using MLFP, Statistical and Actual Value Page

13 LIST OF TABLES Table No Description Performance of Neural Network Model versus Loglinear Regression Model Data Input Actual and Predicted Value Added Page xii

14 LIST OF ABBREVIATIONS ANN GDP MLFP GRNN INV EMP TI' EKS VA LVA LZNV LEKS LEMP MIDA VIF Artificial Neural Network Gross Domestic Product Multi-Layer Feedforward Perception Generalized Regression Neural Network Investment Employee Time Export Value Added Log, represents value added (dependent variable) Log, represents investment (independent variable) Log, represents export (independent variable) Log, represents employee (independent variable) Malaysian Industrial Development Authority Variance Inflation Factor xiii

15 CHAPTER ONE INTRODUCTION BACKGROUND Forecasting is extremely complex activity that influences the setting up of organizations and because of this; there is no single method able to satisfy all forecasting needs. Forecasting is the process of estimation of the value of a variable (or a set of variables) at some future in time. Forecasting plays a significant role in strategic planning, since a lot of activities are to be planned in advanced. By using these forecasts, it can consider several strategic options in effective planning to satisfy needs as well as to increase profits. A forecasting exercise is usually carried out in order to provide an aid to decision making and in planning the future. It works on the premise that "if we can predict what the future will be like, we can modify our behavior now to be in a better position, than otherwise would have been, when the future arrives" (J.E Beasley, 1990). With the advent of modern computer technology and information science, sophisticated information systems can be built that can make decisions or predictions based on information contained in readily available data. Such systems are called Learning System and are currently used for the purpose of classification and prediction. The use of Artificial Neural Network (ANN) is a growing field in computer science.

16 Forecasting the growth and evolution of manufacturing industry in Malaysia is a relevant task in planning strategic options to achieve a more developed future Malaysia: a self-sufficient industrial developed nation, complete with an economy that will be eightfold stronger by the year This pertains to the objectives set out in Vision 2020: To evaluate the level of manufacturing industry in order to plan relevant projects. To make strategic decision, planning strategic option to investment, controlling and modeling the industry. To identify research towards the effects of the changes which occur in order to make necessary action in controlling and facing the predicted affects. This is in line with the objective of developing Malaysia into an industrialized country by the year 2020.

17 PROBLEM STATEMENT Throughout the studies conducted, many techniques have been implemented to perform time series forecasting. Several difficulties arose when performing Time Series Forecasting. Eric A. Plummer (2000) discusses the difficulties which include: a limited quantity of data, data contained noise and non-stationarity - where data that do not have the same statistical properties (e.g., mean and variance) at each point in time. Arbabi, Mansur; Fischthal, Scott M. (1995) have listed the limitation of statistical modeling for forecasting including: function is limited to linear, no relationship between data is assumed, fixed model and difficult to handle noisy or incomplete data. R.G Hoptroff (1993) mentioned that Conventional Forecasting methods are almost all based on linear or linearised models such as Auto Regression Moving Average method. The practical success is limited by their linearity and ravenous data requirements. Zukime M. (2004) mentioned that traditional econometric approach such as the regression equation continue to be used for prediction in most areas

18 such as in academic, economic although the growing application of Neural Network to wider scope of various problems encountered. By means of a case study, this project proposes that Neural Network techniques may successfully be extended to include Prediction Model because of two obvious advantages. First, it does not require any assumptions about underlying population distribution and secondly, it is especially useful for systems that are nonlinear (Zukime M. (2004) OBJECTIVE The aim of this project is to build a predictive model to forecast the growth of manufacturing industry in Malaysia using Artificial Neural Network (ANN). This is to ensure the efficiency of management tool for decision making. SCOPE The study and forecasting will be focus on the manufacturing industry in Malaysia. There are many factors that effect the growth of manufacturing industry in Malaysia. The study will focus on the main factors due to the limitation of data. The main factors that involved in this study are capital, labor for the industry, time that represent time trend and export.

19 The study also will investigate the effectiveness of two architectures in ANN namely Multi-Layer Feedforward Perception (MLFP) and Generalised Regression Neural Network (GRNN) and Multi-Regression technique in Statistical approach. To conclude, comparative study will be performed between the results obtains from Neural Network and the results obtains from Statistical approach. ORGANISATION OF STUDY This paper is organized into six chapters. Chapter One - Introduction deals with the statement of problem, the objectives, the scope, and the organization of the paper. Chapter Two - Literature Review consists of the theoretical studies of forecasting, manufacturing industry in Malaysia as well as the review of previous empirical studies that inspired the researcher to conduct the study. Chapter Three - Research Methodology will describe the research methodology used in this study. It includes input and output specification and software used in this study. The data for the variables is using time series data from 1970 to 2004, and is collected from various sources.

20 Chapter Four - Data Mining: Data Analysis and Model Specification will describe the data analysis and model specification. Model specifications begin with statistical model using Eviews and followed by model specification with Multi-Layer Feedforward Perception and Generalised Regression using NeuralTools. Chapter Five - Result and Discussion will discuss the results of the estimation and interpretation of the findings. The results are consistent with the previous empirical studies, which indicates that the result obtains from neural network is better than statistical approach. Chapter Six - Conclusion and Future Works will provide the conclusion, highlights the problems encountered during the study, and state some recommendation to improve this study.

21 CHAPTER TWO LITERATURE REVIEW INTRODUCTION Any organization, whether commercial, non-profit making organization, or governmental, needs to plan its future economic activities. Planning necessarily requires estimation of its future needs and demands. Therefore, forecasting plays an important role in the planning process of any organization. The ability to accurately forecast or predict future conditions and needs is critical to the survival of organizations. These organizations rely on forecasts so that they can allocate resources efficiently, balance workload against the forecasted demand and plan their operations to meet needs, fulfill the requirements and demands that will be placed upon them and the end-users. The process of forecasting is very similar to modeling (R.G Hoptroff, 1993), where the outcome of an unknown variable is predicted from known or controllable variable. If relevant information about past data is available, the relationship between known and unknown variables, or between past and future events, can be derived through statistical analysis. There are several motivations for forecasting and modeling in business and economic. These include:

22 The Short-Term Forecasts - typically, a short-term forecast is for the week or the month ahead, and is used for stock control and monitoring cash flow. Short-term forecasts are usually based on models of current trends and seasonalities in demand. The Medium-Term Forecasts - typically look at the position over the next few months or years to help manage long-term cash utilization and budgeting. Medium-term forecast incorporate independent influent variables into the trendlseason forecast, to take into account such factors as the cyclical nature of the economy, and the effects of different marketing strategies. The Long-Term Forecasts - typically, a long-term forecast forecasting many years ahead and is normally aids long-term strategic decision making and capital investment programming, and is used both in business and in government. These forecasts are the most difficult of all because of the need to quantify the effect of changes in the fundamental structure of the system. MANUFACTURING INDUSTRY IN MALAYSIA Manufacturing is the transformation of raw materials into finished goods for sale, or intermediate processes involving the production or finishing of semi-

23 manufactures. It is a large branch of industry and of secondary production. Some industries, like semiconductor and steel manufacturers use the term fabrication (wikipedia, 2006). Manufacturing industry in Malaysia has undergone rapid expansion since the 1970s and has become the leading edge of Malaysia's economic growth. It now accounts for the largest share of the country's GDP, although primary activities still employ more workers. Growth has been especially notable in industries assembling electronic equipment, electrical machinery, and appliances and those making chemical products and textiles. The main development goal has been the manufacturer of goods for export, with a lesser emphasis on import substitution. One strategy designed by the government to promote manufactured exports has been the establishment of a number of free trade zones, which have provided duty-free access to imported raw materials and semi-finished parts and numerous investment and export incentives. Industrial estates also have been established in less-developed parts of the country to stimulate manufacturing and to balance industrial growth, but manufacturing capacity has remained highly concentrated. Increasingly, development strategy has shifted to the promotion of small and medium industries that manufacture their own parts and acquire technology from more economically developed countries, the aim being to move beyond the stage of assembly-only manufacturing.

24 Turnover growth statistics of the manufacturing industry are an important link in the production chain of the gross domestic product (GDP). The growth of the turnover of the manufacturing industry is an important source of information for the industrial production index. This is one of the most important short-term economic indicators. To certain extent, this indicator gives a first impression of economic growth or economic decline (R.H.M. Van Der Stegen 2005). RELATED WORKS In recent years, neural network techniques have been increasingly used for a wide variety of applications where statistical methods had been traditionally employed (Widrow, et. al., 1994). Neural network techniques, for example, have been applied to problems like economic forecasting, financial modeling, market share prediction, stock market prediction, and risk management. Neural networks are applicable in virtually every situation in which a relationship between the predictor variables (independents, inputs) and predicted variables (dependents, outputs) exits, even when that relationship is very complex and not easy to articulate in the usual terms of "correlations" or "differences between groups".

25 Zukime M. (2004) has presented a comparative case study between neural network and econometric approaches to predict GDP growth in Malaysia using knowledge based economy indicators based on time series data collected from The best network model obtained in the study consists of 4 input units (IS, MP, NPC and PCI), one hidden layer with one hidden unit, and one output unit (real GDP growth). In the study, the learning rate is set to 0.1, the normalized cumulative delta rule with an epoch of 15 is considered appropriate. On the whole, the neural network model has demonstrated a great potential to predict GDP growth using knowledge based economy indicators as compared to the traditional econometric approach. In education as to predict the student's performance, Naik & Ragothaman (2004) have evaluates the ability of three different models--neural networks, logit, and probit to predict A student performance in graduate program3 Thc neural que was used to classify applicants into successful and mar student pools based on undergraduate GPA, G scores, undergraduate major, age and other relevant data, The results of the study show that the neural network model performs as well as the statistical models and is a useful tool in predicthg A student performance. In economic forecasting, Shaaf, M. (2000) has used an artificial intelligence methodology called "neural networks" and quarterly data from 1959 to 1997 to evaluate the use of yield curve for forecasting recessions. The results confirm