![](/images/graphics-bg.png)
Particle swarm based feature selection for improving random forest classification accuracy
Dissertant
al-Wahidi, Bara Khalid Abd al-Razzaq
Thesis advisor
al-Rusan, Thamir
Abu Hamidah, Mahir Abd al-Qadir Abd al-Munim
University
Isra University
Faculty
Faculty of Information Technology
Department
Department Software Engineering
University Country
Jordan
Degree
Master
Degree Date
2019
English Abstract
Iterating over every possible combination of features and building each combination as a decision tree takes massive processing power especially when there aremany features to select from.
The main drawback with using decision tree classifiers is the tendency of the tree to be over fitted to a specific scenario.
The random forest classifier resolves this issue by using randomly selected features as nodes.
The problem with this approach is that it requires more time and computational power to construct the trees.
Researchers have identified this issue and worked on multiple variations of random forest to reduce the number of decision trees to be grown.
Some of the successful variations use Symmetrical Uncertainty, and other methods to select a feature combination that will yield the highest accuracy achieving trees and generate a random forest for these features rather than the entire dataset.
Others have employed the genetic algorithm in accordance with random forests to optimize the order and appearance of the features in making the random forest.
In this research we employed an optimization algorithm called Binary Particle Swarm.
The binary particle swarm optimization algorithm is a powerful algorithm in the field of optimization.
We used this algorithm to pick the best features that represent a dataset as input for a random forest classifier.
We have achieved impeccable results in terms of accuracy and precision while maintaining minimum user interaction.
We used the Wisconsin breast cancer dataset which can be obtained from the UCI machine learning repository.
The objective in this dataset is to predict whether the patient has a benign or malignant tumor based on the attributes provided.
The other dataset we used was the Titanic disaster dataset which can also be obtained from the UCI machine learning repository.
In this dataset, the objective is to predict whether the passenger has survived or not based on the provided attributes.
We obtained a 97% on average and a best 98% classification accuracy on the Wisconsin breast cancer dataset.
Using the same technique, we obtained 97% classification accuracy on the Titanic datase
Main Subjects
Information Technology and Computer Science
No. of Pages
54
Table of Contents
Table of contents.
Abstract.
Chapter One : Introduction.
Chapter Two : Literature review.
Chapter Three : Related work.
Chapter Four : Proposed approach.
Chapter Five : Implementation.
Chapter Six : Results.
Chapter Seven : Validation.
References.
American Psychological Association (APA)
al-Wahidi, Bara Khalid Abd al-Razzaq. (2019). Particle swarm based feature selection for improving random forest classification accuracy. (Master's theses Theses and Dissertations Master). Isra University, Jordan
https://search.emarefa.net/detail/BIM-988682
Modern Language Association (MLA)
al-Wahidi, Bara Khalid Abd al-Razzaq. Particle swarm based feature selection for improving random forest classification accuracy. (Master's theses Theses and Dissertations Master). Isra University. (2019).
https://search.emarefa.net/detail/BIM-988682
American Medical Association (AMA)
al-Wahidi, Bara Khalid Abd al-Razzaq. (2019). Particle swarm based feature selection for improving random forest classification accuracy. (Master's theses Theses and Dissertations Master). Isra University, Jordan
https://search.emarefa.net/detail/BIM-988682
Language
English
Data Type
Arab Theses
Record ID
BIM-988682