發(fā)布時(shí)間：2020-12-28 06:53

　　在加納,摩托車注冊(cè)數(shù)量差不多占機(jī)動(dòng)車注冊(cè)數(shù)量的四分之一。在加納北部農(nóng)村地區(qū),騎摩托車已成為一種常見又便宜的出行方式。近年來,作為擁堵道路下經(jīng)濟(jì)可行的交通模式,摩托車在加納的城市中也越來越流行。摩托車碰撞事故通常發(fā)生在共用道路上,而與其相關(guān)的傷害與死亡是道路交通安全的重要問題,在近些年顯得尤為突出。目前,摩托車碰撞事故在加納的行人死亡原因中排名第二位。因此,有必要對(duì)導(dǎo)致摩托車碰撞事故的因素進(jìn)行研究。摩托車碰撞事故分析在全球范圍內(nèi)是一個(gè)重要的研究領(lǐng)域。而在加納,還沒有關(guān)于摩托車碰撞事故嚴(yán)重性及其影響因素的研究。目前,關(guān)于預(yù)測(cè)摩托車碰撞事故嚴(yán)重性的經(jīng)典統(tǒng)計(jì)模型,相關(guān)文獻(xiàn)較多。傳統(tǒng)的統(tǒng)計(jì)模型有基本的假設(shè)和預(yù)定義關(guān)系,但如果它們不滿足條件,將產(chǎn)生不準(zhǔn)確的結(jié)果。鑒于統(tǒng)計(jì)模型的缺點(diǎn),本文采用基于機(jī)器學(xué)習(xí)的算法來預(yù)測(cè)摩托車碰撞事故嚴(yán)重性。機(jī)器學(xué)習(xí)技術(shù)采用非參數(shù)模型,其沒有預(yù)測(cè)變量和響應(yīng)變量之間的關(guān)系推定。本文對(duì)不同的機(jī)器學(xué)習(xí)算法進(jìn)行比較和評(píng)價(jià)。本文研究的事故數(shù)據(jù)來自加納建筑與道路研究院(BRRI)的國(guó)家道路交通碰撞數(shù)據(jù)庫(kù)中2011至2015年間的摩托車碰撞數(shù)據(jù)。該數(shù)據(jù)被劃分為4種損傷嚴(yán)重性類型:致命,... 

【文章來源】：江蘇大學(xué)江蘇省

【文章頁(yè)數(shù)】：140 頁(yè)

【學(xué)位級(jí)別】：博士

【文章目錄】：
摘要
ABSTRACT
1 INTRODUCTION
    1.1 Background
    1.2 Problem Statement
    1.3 Research Objectives and Scope
        1.3.1 Research Objective
        1.3.2 Scope of the Research
    1.4 Significance of the Study
    1.5 Structure of the Dissertation
2 REVIEW OF FACTORS AFFECTING MOTORCYCLE CRASH SEVERITY AND MOTORCYCLE CRASH SEVERITY ANALYSIS METHODS
    2.1 Introduction
    2.2 Contributing Factors Responsible to Motorcycle Crash Severity
        2.2.1 Motorcyclists’Characteristics
        2.2.2 Roadway Features and Roadside Fittings
        2.2.3 Crash Characteristics
        2.2.4 Temporal Characteristics
        2.2.5 Environmental Conditions
    2.3 Crash Severity Studies
        2.3.1 Characteristics of crash-injury severity data
        2.3.2 Traditional Statistical Techniques for Motorcycle Crash Severity Analysis
        2.3.3 Machine Learning Techniques for Motorcycle Crash Severity Analysis
    2.4 Summary
3 PROPOSED RESEARCH METHODS
    3.1 Introduction
    3.2 Classification Methods
        3.2.1 Neural Networks
        3.2.2 Rule-based classification
        3.2.3 Classification and Regression Trees
        3.2.4 J48 decision tree classifier
        3.2.5 Instance-Based learning with parameter k
    3.3 Ensemble methods
        3.3.1 Introduction
        3.3.2 AdaBoost
        3.3.3 Bagging
        3.3.4 Random Forest
        3.3.5 Majority Vote Combiner
    3.4 Modeling Tools
    3.5 Validation of the models
    3.6 Performance metrics
    3.7 Quantifying the Contributing Factors of Motorcyclist Injury Severity
    3.8 Summary
4 PREPARATION AND UNDERSTANDING OF DATA
    4.1 Introduction
    4.2 Data Collection
    4.3 Data Processing
    4.4 Description of the Data
    4.5 Summary
5 CONFIGURATION OF MACHINE LEARNING MODELS FOR PREDICTION OF MOTORCYCLE CRASH SEVERITY
    5.1 Introduction
    5.2 Loading the data into the WEKA
    5.3 Classifiers
        5.3.1 Multilayer Perceptron
        5.3.2 PART:Rule-based classifier
        5.3.3 Classification and Regression Trees
        5.3.4 J48 decision tree classifier
        5.3.5 Instance-Based learning with parameter k
    5.4 Improving Results with Construction of Ensembles
    5.5 Quantifying the Contributing Factors of Motorcyclist Injury Severity
    5.6 Summary
6 RESULTS AND COMPARATIVE ANALYSIS OF DEVELOPED CLASSIFIERS
    6.1 Introduction
    6.2 Comparing Results of Ensembles and Individual Classifiers
        6.2.1 Individual Classifiers
        6.2.2 Classifier Ensemble
    6.3 Quantifying the Contributing Factors of Motorcyclist Injury Severity
7 CONCLUSIONS AND FUTURE RESEARCH DIRECTIONS
    7.1 Introduction
    7.2 Conclusions
    7.3 Future Research Directions
REFERENCE
ACKNOWLEDGEMENTS
PUBLICATIONS
Appendix A Summary of Output from the Classifiers
    Appendix A.1 MLP Classifier
    Appendix A.2 PART Classifier
    Appendix A.3 CART Classifier
    Appendix A.4 J48 Classifier
    Appendix A.5 IBk Classifier
Appendix B Summary of Output from the Ensembles
    Appendix B.1 AdaBoost
        Appendix B1.1 BoostingMLP
        Appendix B1.2 BoostingPART
        Appendix B1.3 BoostingCART
        Appendix B1.4 BoostingJ
        Appendix B1.5 BoostingIBk
    Appendix B.2 Bagging
        Appendix B2.1 BaggingMLP
        Appendix B2.2 BaggingPART
        Appendix B2.3 BaggingCART
        Appendix B2.4 BaggingJ
        Appendix B2.5 BaggingIBk
    Appendix B.3 Random Forest
    Appendix B.4 Majority Vote Combiner
Appendix C Summary of Output from the Evaluator



本文編號(hào)：2943320