Quality of Data in Machine Learning

TL;DR

This paper challenges the common belief that more data always improves machine learning performance, showing that data quality is more crucial than quantity, based on empirical experiments with vocational student data.

Contribution

It provides empirical evidence that increasing data quantity does not necessarily enhance model accuracy, emphasizing the importance of data quality over quantity.

Findings

Increasing data records does not significantly improve accuracy.

Variance of model accuracy decreases with ensemble models.

Data quality is more important than data quantity for model performance.

Abstract

A common assumption exists according to which machine learning models improve their performance when they have more data to learn from. In this study, the authors wished to clarify the dilemma by performing an empirical experiment utilizing novel vocational student data. The experiment compared different machine learning algorithms while varying the number of data and feature combinations available for training and testing the models. The experiment revealed that the increase of data records or their sample frequency does not immediately lead to significant increases in the model accuracies or performance, however the variance of accuracies does diminish in the case of ensemble models. Similar phenomenon was witnessed while increasing the number of input features for the models. The study refutes the starting assumption and continues to state that in this case the significance in data lies in the quality of the data instead of the quantity of the data.