ICET Presentation 1

download ICET Presentation 1

of 23

Transcript of ICET Presentation 1

  • 7/31/2019 ICET Presentation 1

    1/23

    A Feature Subset Selection Methodbased on Ant Colony Optimization

    and Symmetric Uncertainty

    Authors: Syed Imran Ali*,Dr. Wasem Shahzad. (NU-FAST)

    * Presenter

  • 7/31/2019 ICET Presentation 1

    2/23

    Presentation Layout

    Introduction

    Motivation

    Background on Feature Selection

    Proposed Technique

    Ant Colony Optimization

    Symmetric Uncertainty Experimentation

    Conclusion

    2

  • 7/31/2019 ICET Presentation 1

    3/23

    Motivation

    Why Feature Selection?

    Curse of Dimensionality

    Three-fold benefits

    Enhance the capabilities of Filter based methods

    3

  • 7/31/2019 ICET Presentation 1

    4/23

    Data Reduction

    4

  • 7/31/2019 ICET Presentation 1

    5/23

    Feature Selection Types

    5

  • 7/31/2019 ICET Presentation 1

    6/23

    PROPOSED TECHNIQUE

  • 7/31/2019 ICET Presentation 1

    7/23

    Proposed Technique

    Basic Ingredients of ACO Graph Representation

    Heuristic Desirability

    Positive feedback process

    Constraint Satisfaction Solution Construction mechanism

    7

  • 7/31/2019 ICET Presentation 1

    8/23

    Basic Ingredients of ACO

    Graph Representation

    8

  • 7/31/2019 ICET Presentation 1

    9/23

    Heuristic desirability and PositiveFeedback mechanism

    Basic Ingredients of ACO

    9

  • 7/31/2019 ICET Presentation 1

    10/23

    Constraint Satisfaction and SolutionConstruction

    Basic Ingredients of ACO

    10

  • 7/31/2019 ICET Presentation 1

    11/23

    Information Theoretic Measure

    Information Gain

    Symmetric Uncertainty

    IG (Y,X) = H(Y) H (Y|X)

    11

  • 7/31/2019 ICET Presentation 1

    12/23

    ACO-SU

    12

  • 7/31/2019 ICET Presentation 1

    13/23

    EXPERIMENTATION

  • 7/31/2019 ICET Presentation 1

    14/23

    Experimentation Framework

    All the experiments are performed using 10-FoldCross Validation and results of ten runs are

    averaged.

    Proposed method is compared with four otherfeature selection algorithms.

    Performance Metrics:

    Number of Features Selected.

    Predictive Classification Accuracy using 10-FCV.

    14

  • 7/31/2019 ICET Presentation 1

    15/23

    Experimentation Framework

    SNO. Dataset Total Features Instances Classes1 * Iris 4 150 32 Liver Disorder 6 345 23 Diabetes 8 768 24 Breast Cancer- W 9 699 25 Vote 16 435 26 Labor 16 57 27 Hepatitis 19 155 28 Colic-Horse 22 368 29 Ionosphere 34 351 210 Lymph 18 148 411 Dermatology 34 366 612 Lung Cancer 56 32 313 Audiology 69 226 24

    15

    6

  • 7/31/2019 ICET Presentation 1

    16/23

    Experimentation Framework

    Parameters Values

    Number of Ants 20

    Alpha 1

    Beta 1

    Evaporation Rate 0.15

    Max. Epochs 500

    Path Convergence Threshold 50

    16

  • 7/31/2019 ICET Presentation 1

    17/23

    Experiment

    17

    8

  • 7/31/2019 ICET Presentation 1

    18/23

    Experiment

    18

  • 7/31/2019 ICET Presentation 1

    19/23

    Experiment

    19

    20

  • 7/31/2019 ICET Presentation 1

    20/23

    Features Selected by ACO-SU

    Dataset Total ACO-SUIris 4 2Liver Disorder 6 2Diabetes 8 2Breast Cancer- W 9 4Vote 16 6Labor 16 6Hepatitis 19 7Colic-Horse 22 6Ionosphere 34 9Lymph 18 7Audiology 69 20Dermatology 34 12Lung Cancer 32 25

    20

    21

  • 7/31/2019 ICET Presentation 1

    21/23

    No. of Features selected

    21

    22

  • 7/31/2019 ICET Presentation 1

    22/23

    Conclusion We have proposed an efficient feature selection

    method based on SI and filter method techniques.

    Proposed method is extensively experimented over anumber of benchmark datasets and classifiers.

    ACO-SU yields better results as compared to otherSI based feature selection methods considered in the

    study. ACO-SU outperformed other methods in terms of

    predictive classification accuracy and number offeatures selected.

    22

    23

  • 7/31/2019 ICET Presentation 1

    23/23

    Thank You

    Questions?

    23