Classification of highly unbalanced data using deep learning techniques

Slides:



Advertisements
Similar presentations
ECG Signal processing (2)
Advertisements

Classification and Decision Boundaries
Ensemble Learning what is an ensemble? why use an ensemble?
CS 590M Fall 2001: Security Issues in Data Mining Lecture 3: Classification.
CS 8751 ML & KDDEvaluating Hypotheses1 Sample error, true error Confidence intervals for observed hypothesis error Estimators Binomial distribution, Normal.
Fraud Detection Experiments Chase Credit Card –500,000 records spanning one year –Evenly distributed –20% fraud, 80% non fraud First Union Credit Card.
Visual Categorization with Bag of Keypoints
Face Processing System Presented by: Harvest Jang Group meeting Fall 2002.
Predicting Income from Census Data using Multiple Classifiers Presented By: Arghya Kusum Das Arnab Ganguly Manohar Karki Saikat Basu Subhajit Sidhanta.
Cost-Sensitive Bayesian Network algorithm Introduction: Machine learning algorithms are becoming an increasingly important area for research and application.
Machine Learning1 Machine Learning: Summary Greg Grudic CSCI-4830.
1 1 Slide Evaluation. 2 2 n Interactive decision tree construction Load segmentchallenge.arff; look at dataset Load segmentchallenge.arff; look at dataset.
TOPICS IN BUSINESS INTELLIGENCE K-NN & Naive Bayes – GROUP 1 Isabel van der Lijke Nathan Bok Gökhan Korkmaz.
Ensemble Methods: Bagging and Boosting
Introduction Use machine learning and various classifying techniques to be able to create an algorithm that can decipher between spam and ham s. .
Reservoir Uncertainty Assessment Using Machine Learning Techniques Authors: Jincong He Department of Energy Resources Engineering AbstractIntroduction.
A Brief Introduction and Issues on the Classification Problem Jin Mao Postdoc, School of Information, University of Arizona Sept 18, 2015.
Evaluating Classifiers Reading: T. Fawcett, An introduction to ROC analysis, Sections 1-4, 7 (linked from class website)An introduction to ROC analysis.
Deep Residual Learning for Image Recognition
Feasibility of Using Machine Learning Algorithms to Determine Future Price Points of Stocks By: Alexander Dumont.
Classification of Breast Cancer Cells Using Artificial Neural Networks and Support Vector Machines Emmanuel Contreras Guzman.
Lecture 3a Analysis of training of NN
Neural networks and support vector machines
Batch Normalization: Accelerating Deep Network Training by Reducing Internal Covariate Shift CS838.
Intrusion Detection using Deep Neural Networks
Results for all features Results for the reduced set of features
Prepared by: Mahmoud Rafeek Al-Farra
Can-CSC-GBE: Developing Cost-sensitive Classifier with Gentleboost Ensemble for breast cancer classification using protein amino acids and imbalanced data.
Understanding Neural Networks using .NET
Evaluating Results of Learning
Performance Measures II
Feature Selection for Pattern Recognition
Cost-Sensitive Learning
Schizophrenia Classification Using
Evaluating classifiers for disease gene discovery
CS 698 | Current Topics in Data Science
Generalization ..
Data Mining Classification: Alternative Techniques
Deep Learning Convoluted Neural Networks Part 2 11/13/
Machine Learning Week 1.
Students: Meiling He Advisor: Prof. Brain Armstrong
Machine Learning Today: Reading: Maria Florina Balcan
Improved Rooftop Detection in Aerial Images with Machine Learning
SEG 4630 E-Commerce Data Mining — Final Review —
Cost-Sensitive Learning
Prepared by: Mahmoud Rafeek Al-Farra
ECE 599/692 – Deep Learning Lecture 4 – CNN: Practical Issues
Classification of class-imbalanced data
COSC 4335: Other Classification Techniques
Neural Networks II Chen Gao Virginia Tech ECE-5424G / CS-5824
Model generalization Brief summary of methods
Data Mining Class Imbalance
ImageNet Classification with Deep Convolutional Neural Networks
Basics of ML Rohan Suri.
Deep Residual Learning for Automatic Seizure Detection
Machine Learning with Clinical Data
Naive Bayes Classifier
Neural Networks II Chen Gao Virginia Tech ECE-5424G / CS-5824
Identifying Severe Weather Radar Characteristics
Assignment 1: Classification by K Nearest Neighbors (KNN) technique
A task of induction to find patterns
MAS 622J Course Project Classification of Affective States - GP Semi-Supervised Learning, SVM and kNN Hyungil Ahn
Review Session II.
Introduction to Neural Networks
Review Session II.
A task of induction to find patterns
Mastering Open-face Chinese Poker by Self-play Reinforcement Learning
COSC 4368 Intro Supervised Learning Organization
LHC beam mode classification
Credit Card Fraudulent Transaction Detection
Presentation transcript:

Classification of highly unbalanced data using deep learning techniques Dianjing Liu

Thyroid Disease Data Set (Sick) (Not Sick) Negative: 3178 Positive: 250 Predicting all instances to be negative: 92.71% accuracy, 0% sensitivity. Testing accuracy of previous classifiers: Nearest-neighbor: 92.44% k-Nearest Neighbor: 93.70% Ref. [1]: 93.23% Bayes rule: 96.1% Nonlinear Bayes: 97.2% Ref. [2]: 97.23% Ref. [3]: 99.36%

Neural network settings Activation: sigmoid Training steps: 50,000 Learning rate: 0.002. Decays by half every 5,000 steps. Network size is chosen by cross validation:

Oversampling Original Oversampled Table 1. The accuracy and sensitivity of neural networks trained with different positive instance ratios.

Oversampling Original Oversampled ROC curve PR curve

Other techniques Activation function Dropout Batch normalization Adam Optimization algorithm

Other techniques Learning curves: Tabel 2. Testing results of different techniques (with or without oversampling). The ReLU activation and dropout methods fail to train effective models and predict all samples to be negative. Learning curves: Oversampling BN and Adam

Conclusion Oversampling can always improve sensitivity. BN and Adam have very good performance: higher accuracy, sensitivity and training speed. Best model: 98.60% accuracy. 99.60% sensitivity.