Low Classification Accuracy by Logistic Regression, Support Vector Classifier, and Multi Layer Perceptron, but Not Decision Tree, on Random Attributes from Hadamard Matrix.

Citation: Ling, MHT. 2020. Low Classification Accuracy by Logistic Regression, Support Vector Classifier, and Multi-Layer Perceptron, but Not Decision Tree, on Random Attributes from Hadamard Matrix. EC Clinical and Medical Case Reports 3(12): 07-10.

Link to [abstract] and [PDF].

Here is the permanent [PDF] and [data set] links to my archive.

The use of machine learning classifiers is increasing with evidence of overtaking human judgement. This can be risky if workings and implications of machine learning classifiers remain a black box. Here, a case where a balanced and algorithmically generated data set, Hadamard matrix, classifies poorer than random using logistic regression (accuracy < 17.4%), support vector classifier (accuracy < 23.4%) and in most cases of multi-layer perceptron (accuracy < 27.9%) but not in decision tree (accuracy > 77.3%); despite perfect (100%) internal classification accuracy for both support vector classifier and multi-layer perceptron; is reported. This suggests a systematic and yet currently unexplained source of error.