RapidMiner. Business Analytics Applications. Data Mining Use Cases and. Markus Hofmann. Ralf Klinkenberg. Rapid-I / RapidMiner.

Transcription

1 RapidMiner Data Mining Use Cases and Business Analytics Applications Edited by Markus Hofmann Institute of Technology Blanchardstown, Dublin, Ireland Ralf Klinkenberg Rapid-I / RapidMiner Dortmund, Germany CRC Press Taylor& Francis Croup Boca Raton London New York CRC Press is an imprint of the Taylor S Francis Group, an informa business A CHAPMAN & HALL BOOK

2 Contents I Introduction to Data Mining and RapidMiner 1 1 What This Book is About and What It is Not 3 Ingo Mierswa 1.1 Introduction Coincidence or Not? Applications of Data Mining Financial Services Retail and Consumer Products Telecommunications and Media Manufacturing, Construction, and Electronics Fundamental Terms Attributes and Target Attributes Concepts and Examples Attribute Roles Value Types Data and Meta Data Modeling 16 2 Getting Used to RapidMiner 19 Ingo Mierswa 2.1 Introduction First Start Design Perspective Building a First Process Loading Data Creating a Predictive Model Executing a Process Looking at Results 29 II Basic Classification Use Cases for Credit Approval and in Education 31 3 k-nearest Neighbor Classification I 33 M. Farced Akhtar 3.1 Introduction Algorithm The k-nn Operator in RapidMiner Dataset Teacher Assistant Evaluation Dataset Basic Information Examples 35 ix

3 x Contents Attributes Operators in This Use Case Read URL Operator Rename Operator Numerical to Binominal Operator Numerical to Polynominal Operator Set Role Operator Split Validation Operator Apply Model Operator Performance Operator Use Case Data Import Pre-processing Renaming Attributes Changing the Type of Attributes Changing the Role of Attributes Model Training, Testing, and Performance Evaluation 41 4 k-nearest Neighbor Classification II 45 M. Fareed Akhtar 4.1 Introduction Dataset Operators Used in This Use Case Read CSV Operator Principal Component Analysis Operator Split Data Operator Performance (Classification) Operator Data Import Pre-processing Principal Component Analysis Model Training, Testing, and Performance Evaluation Training the Model Testing the Model Performance Evaluation 51 5 Naive Bayes Classification I 53 M. Fareed Akhtar 5.1 Introduction Dataset Credit Approval Dataset Examples Attributes Operators in This Use Case Rename by Replacing Operator Filter Examples Operator Discretize by Binning Operator X-Validation Operator Performance (Binominal Classification) Operator Use Case Data Import Pre-processing 58

4 Contents xi Model Training, Testing, and Performance Evaluation 61 6 Naive Bayes Classificaton II 65 M. Fareed Akhtar 6.1 Dataset Nursery Dataset Basic Information Examples Attributes Operators in this Use Case Read Excel Operator Select Attributes Operator Use Case Data Import Pre-processing Model Training, Testing, and Performance Evaluation A Deeper Look into the Naive Bayes Algorithm 71 III Marketing, Cross-Selling, and Recommender System Use Cases 75 7 Who Wants My Product? Affinity-Based Marketing 77 Euler Timm 7.1 Introduction Business Understanding Data Understanding Data Preparation Assembling the Data Preparing for Data Mining Modelling and Evaluation Continuous Evaluation and Cross Validation Class Imbalance Simple Model Evaluation Confidence Values, ROC, and Lift Charts Trying Different Models Deployment Conclusions 94 8 Basic Association Rule Mining in RapidMiner 97 Matthew A. North 8.1 Data Mining Case Study 97 9 Constructing Recommender Systems in RapidMiner 119 Matej Mihelcic, Matko Bosnjak, Nino Antulov-Fantulin, and Tomislav Smuc 9.1 Introduction The Recommender Extension Recommendation Operators Data Format Performance Measures The VideoLectures.net Dataset Collaborative-based Systems 127

5 xii Contents Neighbourhood-based Recommender Systems Factorization-based Recommender Systems Collaborative Recommender Workflows Iterative Online Updates Content-based Recommendation Attribute-based Content Recommendation Similarity-based Content Recommendation Hybrid Recommender Systems Providing RapidMiner Recommender System Workflows as Web Services Using R.apidAnalytics Simple Recommender System Web Service Guidelines for Optimizing Workflows for Service Usage Summary Recommender System for Selection of the Right Study Program for Higher Education Students 145 Milan Vukicevic, Milos Jovanovic, Boris Delibasic, and Milija Suknovic 10.1 Introduction Literature Review Automatic Classification of Students using RapidMiner Data Processes Simple Evaluation Process Complex Process (with Feature Selection) Results Conclusion 155 IV Clustering in Medical and Educational Domains Visualising Clustering Validity Measures 159 Andrew Chisholm 11.1 Overview Clustering A Brief Explanation of k-means Cluster Validity Measures Internal Validity Measures External Validity Measures Relative Validity Measures The Data Artificial Data E-coli Data Setup Download and Install R. Extension Processes and Data The Process in Detail Import Data (A) Generate Clusters (B) Generate Ground Truth Validity Measures (C) Generate External Validity Measures (D) Generate Internal Validity Measures (E) Output Results (F) 174

6 Contents xiii 11.7 Running the Process and Displaying Results Results and Interpretation Artificial Data E-coli Data Conclusion Grouping Higher Education Students with RapidMiner 185 Milan Vukicevic, Milos Jovanovic, Boris Delibasic, and Milija Suknovic 12.1 Introduction Related Work Using RapidMiner for Clustering Higher Education Students Data Process for Automatic Evaluation of Clustering Algorithms Results and Discussion Conclusion 193 V Text Mining: Spam Detection, Language Detection, and Customer Feedback Analysis Detecting Text Message Spam 199 Neil McGuigan 13.1 Overview Applying This Technique in Other Domains Installing the Text Processing Extension Getting the Data Loading the Text Data Import Wizard Step Data Import, Wizard Step Data Import Wizard Step Data Import Wizard Step Step Examining the Text Tokenizing the Document Creating the Word List and Word Vector Examining the Word Vector Processing the Text for Classification Text Processing Concepts The Naive Bayes Algorithm How It Works Classifying the Data as Spam or Ham Validating the Model Applying the Model to New Data Running the Model on New Data Improvements Summary Robust Language Identification with RapidMiner: A Text Mining Use Case 213 Matko Bosnjak, Eduarda Mendes Rodrigues, and Luis Sarmento 14.1 Introduction The Problem of Language Identification 215

7 xiv Contents 14.3 Text Representation Encoding Token-based Representation Character-Based Representation Bag-of-Words Representation Classification Models Implementation in RapidMiner Datasets Importing Data Frequent Words Model Character n-grams Model Similarity-based Approach Application Rapid Analytics Web Page Language Identification Summary Text Mining with RapidMiner 241 Gurdal Ertek, Dilek Tapucu, and Inane Arin 15.1 Introduction Text Mining Data Description Running RapidMiner RapidMiner Text Processing Extension Package Installing Text Mining Extensions Association Mining of Text Document Collection (ProcessOl) Importing ProcessOl Operators in ProcessOl Saving ProcessOl Clustering Text Documents (Process02) Importing Process Operators in Process Saving Process Running ProcessOl and Analyzing the Results Running ProcessOl Empty Results for ProcessOl Specifying the Source Data for ProcessOl Re-Running ProcessOl ProcessOl Results Saving ProcessOl Results Running Process02 and Analyzing the Results Running Process Specifying the Source Data for Process Process02 Results 257

8 Contents xv 15.6 Conclusions 261 VI Feature Selection and Classification in Astroparticle Physics and in Medical Domains Application of RapidMiner in Neutrino Astronomy 265 Tim Ruhe, Katharina Morik, and Wolfgang Rhode 16.1 Protons, Photons, and Neutrinos Neutrino Astronomy Feature Selection Installation of the Feature Selection Extension for RapidMiner Feature Selection Setup Inner Process of the Loop Parameters Operator Inner Operators of the Wrapper X-Validation Settings of the Loop Parameters Operator Feature Selection Stability Event Selection Using a Random Forest The Training Setup The Random Forest in Greater Detail The Random Forest Settings The Testing Setup Summary and Outlook Medical Data Mining 289 Mertik Matej and Palfy Miroslav 17.1 Background Description of Problem Domain: Two Medical Examples Carpal Tunnel Syndrome Diabetes Data Mining Algorithms in Medicine Predictive Data Mining Descriptive Data Mining Data Mining and Statistics: Hypothesis Testing Knowledge Discovery Process in RapidMiner: Carpal Tunnel Syndrome Defining the Problem, Setting the Goals Dataset Representation Data Preparation Modeling Selecting Appropriate Methods for Classification Results and Data Visualisation Interpretation of the Results Hypothesis Testing and Statistical Analysis Results and Visualisation Knowledge Discovery Process in RapidMiner: Diabetes Problem Definition, Setting the Goals Data Preparation Modeling Results and Data Visualization Hypothesis Testing Specifics in Medical Data Mining Summary 316

9 xvi Contents VII Molecular Structure- and Property-Activity Relationship Modeling in Biochemistry and Medicine Using PaDEL to Calculate Molecular Properties and Chemoinformatic Models 321 Markus Muehlbacher and Johannes Kornhuber 18.1 Introduction Molecular Structure Formats for Chemoinformatics Installation of the PaDEL Extension for RapidMiner Applications and Capabilities of the PaDEL Extension Examples of Computer-aided Predictions Calculation of Molecular Properties Generation of a Linear Regression Model Example Workflow Summary Chemoinformatics: Structure- and Property-activity Relationship Devel opment 331 Markus Muehlbacher and Johannes Kornhuber 19.1 Introduction Example Workflow Importing the Example Set Preprocessing of the Data Feature Selection Model Generation Validation Y-Randomization Results Conclusion/Summary 340 VIII Image Mining: Feature Extraction, Segmentation, and Classification Image Mining Extension for RapidMiner (Introductory) 347 Radim Burget, Vaclav Uher, and Jan Masek 20.1 Introduction Image Reading/Writing Conversion between Colour and Grayscale Images Feature Extraction Local Level Feature Extraction Segment-Level Feature Extraction Global-Level Feature Extraction Summary Image Mining Extension for RapidMiner (Advanced) 363 Vaclav Uher and B.adim Burget 21.1 Introduction Image Classification Load Images and Assign Labels Global Feature Extraction Pattern Detection 368

10 Contents xvii Process Creation Image Segmentation and Feature Extraction Summary 373 IX Anomaly Detection, Instance Selection, and Prototype Construction Instance Selection in RapidMiner 377 Marvin Blachnik and Miroslaw Kordos 22.1 Introduction Instance Selection and Prototype-Based Rule Extension Instance Selection Description of the Implemented Algorithms Accelerating 1-NN Classification Outlier Elimination and Noise Reduction Advances in Instance Selection Prototype Construction Methods Mining Large Datasets Summary Anomaly Detection 409 Markus Goldstein 23.1 Introduction Categorizing an Anomaly Detection Problem Type of Anomaly Detection Problem (Pre-processing) Local versus Global Problems Availability of Labels A Simple Artificial Unsupervised Anomaly Detection Example Unsupervised Anomaly Detection Algorithms k-nn Global Anomaly Score Local Outlier Factor (LOF) Connectivity-Based Outlier Factor (COF) Influenced Outlierness (INFLO) Local Outlier Probability (LoOP) Local Correlation Integral (LOCI) and aloci Cluster-Based Local Outlier Factor (CBLOF) Local Density Cluster-Based Outlier Factor (LDCOF) An Advanced Unsupervised Anomaly Detection Example Semi-supervised Anomaly Detection Using a One-Class Support Vector Machine (SVM) Clustering and Distance Computations for Detecting Anomalies Summary 433 X Meta-Learning, Automated Learner Selection, Feature Selection, and Parameter Optimization Using RapidMiner for Research: Experimental Evaluation of Learners 439 Jovanovic Milos, Vukicevic Milan, Delibasic Boris, and Suknovic Milija 24.1 Introduction Research of Learning Algorithms Sources of Variation and Control 440

11 xviii Contents Example of an Experimental Setup Experimental Evaluation in RapidMiner Setting Up the Evaluation Scheme Looping Through a Collection of Datasets Looping Through a Collection of Learning Algorithms Logging and Visualizing the Results Statistical Analysis of the R.esults Exception Handling and Parallelization Setup for Meta-Learning Conclusions 452 Subject Index 455 Operator Index 463