Machine Learning Versus Statistics
A study published last week used a wave flume to evaluate the effects of waves on small-diameter pipelines. The experimental data was interpreted statistically and using machine learning.
The difference between the two is fundamental.
As a University of North Dakota blog pointed out recently: In statistics, the main objective is to understand relationships between variables, make predictions, test scientific hypotheses and provide explanations based on data. The focus is often on understanding the underlying processes that generate the data and providing interpretable results that can inform decision-making.
In contrast, machine learning aims to develop algorithms that can learn from data and make accurate predictions or decisions without being explicitly programmed. The focus is on the model's performance using large datasets in practical applications, such as classifying images, recognizing speech or predicting behavior.
The authors of the pipeline study, from Kuwait Institute for Scientific Research, compared four machine learning systems with statistical regression analysis and found that the machine learning provided better predictive performance (lower Mean Squared Error values across most target variables).
Machine learning models, such as Extreme Gradient Boosting (XGBoost), can learn from experimental data and capture hidden dependencies between pipeline geometry, seabed conditions and hydrodynamic forces without relying on predefined force coefficients, the researchers say. Unlike empirical methods, machine learning models can generalize predictions for untested conditions, making them valuable tools for engineering design and stability assessments.
“On the other hand, the Gamma Regression models, while slightly less accurate, offered far greater interpretability. They provided explicit equations and coefficients that clarified how each predictor variable influenced the target outcomes, facilitating deeper insights into the physical processes at play.”
Machine learning models have been used in ocean engineering to calculate the equilibrium scour depth around piles, to predict wave formation, to simulate coastal processes, to predict iceberg draft and to predict changes in beach profiles.