Enhancing Random Forest Classifier using Genetic Algorithm

Sania Jawaid, Mohd Abdul Ahad

Abstract


Classification is a problem of distinguishing and categorizing an observation into sub-populations or groups on the basis of certain prior observations whose category membership is known which are called the training data sets. Classification can also be termed as a part of pattern recognition. Its usage is not only limited to the computer analysis from discerning spam emails from genuine ones but also in daily use by doctors for classifying diseases of the patients by observing their characteristics e.g. blood pressure, heart rate, symptoms etc. Therefore, it becomes highly important to have accurate classification methods so that problems and its causes can be identified quickly and accurately in order to solve them. One such algorithm for classification in machine learning and statistics is the classifier called ‘Random Forest’. While decision trees lack behind with their low bias and high variance trade-off, Random Forest is one of the algorithms in supervised learning where low bias and comparatively lower variance than decision trees win for large training data sets as they have low asymptotic error. However, by reducing the correlation between trees we can further reduce the variance and hence improve the algorithm. Therefore, by modifying the existing algorithm by overcoming a few of its demerits can make a classifier more accurate and trustworthy. This paper tries to propose a solution by combining one of the optimizing strategies i.e. Genetic Algorithm with Random Forest to overcome its problem of over-fitting the datasets.

Keywords


Machine learning; supervised learning; Classifier; Random Forests; over-fitting; Genetic Algorithm

Full Text:

PDF


DOI: https://doi.org/10.26483/ijarcs.v8i5.3358

Refbacks

  • There are currently no refbacks.




Copyright (c) 2017 International Journal of Advanced Research in Computer Science