A Comparative Study Of Machine Learning Algorithms For Predictive Data Analytics

Abstract

Machine learning technologies have found numerous applications in predictive data analytics through making reliable predictions and intelligent decision-making. Decision Tree (DT), Random Forest (RF), Support Vector Machine (SVM), and Artificial Neural Network (ANN) are four popular supervised machine learning algorithms that were compared in this paper to find the best one for predictive data analytics. A quantitative comparative research approach is used in this study involving a predictive benchmark dataset that has been prepared using such data preprocessing procedures as cleaning, normalization, categorical encoding, and splitting into 80% training and 20% testing. Time spent training and time spent making predictions are some of the performance metrics used to evaluate four algorithms. The F1-score, precision, accuracy, and area under the receiver operating characteristic (ROC) curve are among the additional measurements. As a result, the best predictive performance was provided by the Artificial Neural Network algorithm with the following metrics: accuracy – 95.8%; precision – 95.2%; recall – 94.8%; F1-score – 95.0%; and AUC (Area Under the ROC curve) score – 0.98; secondly follows Random Forest with the accuracy – 94.6% and the AUC – 0.97. Although Decision Tree showed the best performance in training and prediction times (respectively, 1.8 s and 5 ms), it is less accurate compared to other algorithms. Thus, according to the research results, the Artificial Neural Network and Random Forest algorithms are the most balanced between their predictive efficiency and reliability, while Decision Trees are still relevant for fast-executing tasks.