ML model optimization with ensemble learning, retraining
This article is excerpted from the course “Fundamental Machine Learning,” part of the Machine Learning Specialist certification program from Arcitura Education. It is the eleventh part of the 13-part series, “Using machine learning algorithms, practices and patterns.”
Model optimization techniques are used to improve the performance of machine learning models. This and the next, final article in this series cover a set of optimization techniques that are normally applied toward the end of a machine learning problem-solving task, after a given model has been trained but when there exist opportunities to make it more effective.
This article describes the first two of four optimization practices: the ensemble learning and the frequent model retraining techniques. As explained in Part 4, these techniques are documented in a standard pattern profile format.
Ensemble learning: Overview
How can the accuracy of a prediction task be increased when different prediction models provide varying levels of accuracy?
Developing different models either using the same algorithm with varied training data or using different algorithms with the same training data often results in varied level of model accuracy, which makes using a particular model less than optimal for solving a machine learning task.
Multiple models are built and used by intelligently combining the results of the models in such a way that the resulting accuracy is higher than any individual model alone.
Either homogeneous or heterogeneous models (for classification or regression) are developed. Techniques such as bagging, boosting or random forests are then employed to create a meta-model.