Comparison of Random Forest, Neural Network and Naïve Baye’s in Classification Problem– an Illustration using Orange

Abstract

In this study, we use the orange data mining platform to examine the performance of popular classification algorithms: Random Forest, Neural Network, and Naïve Bayes. Assessing and comparing these models' precision, interpretability, and computational effectiveness on a standard classification task is the aim. The tests were conducted using a classical iris dataset comprising of 150 instances of iris setosa, iris virginica and iris versicolor. Models were trained and evaluated using 10-fold cross-validation using Orange's visual programming environment, and performance metrics such as recall, accuracy, precision, F1 score, and AUC were noted. The findings show that Random Forest continuously produced the best accuracy and robustness, Neural Networks performed competitively but needed to be carefully adjusted to prevent overfitting, and Naïve Bayes provided ease of use and quick training times but had somewhat worse predictive accuracy. Orange uses drag and drop techniques and assist in the implementation of model which otherwise required coding and expertise in programming language. The insights are useful in identifying the appropriate method and choose the best algorithms.