Hind LBAHY
Automated ML is a set of methods that allow non-experts to
use Machine Learning so easily, to ameliorate its efficiency and to dig more in their researches. Machine learning knew recently a big success and an ever-growing number of fields rely on its performance. Nevertheless, this success is based on human machine learning experts to do some tasks manually.
Moreover, because these tasks are complex, especially for non-
ML-experts, a rapid growth of machine learning applications has appeared and the demand for ML methods and processes that is handy and simple to use is more and more important.
The design of an effective ML requires expert’s knowledge to
improve algorithms in order
to optimize results. So how can we use, for example, K-fold cross-validation to look for an optimal tuning parameter? And
how can this process be made more efficient? The idea of machine learning pipeline is based on the automation of machine
learning workflows. It consists on training a given model
through several steps. However, we can remark that the word ‘pipeline’ refers to one-way flow of data which is not the case of
ML pipelines. In this context, machine learning pipelines rely on repetition of steps and continuous improvement of learning which make them very cyclical and iterative. Therefore, we get successful algorithms and models with good accuracy. The development of the Tree-based Pipeline Optimization Tool (TPOT) recently allows the optimization and the automatic design of the machine learning pipelines when given a problem to deal with without human participation. Therefore, how can we optimize machine learning pipelines using TPOT and a version of genetic programming? which is an automated method for creation of computer programs from a high-level problem statement of a problem.To see how to
optimize the automatic design of machine learning pipelines, we will put all these algorithms into practical application to
manage the information system incidents of one the insurance leaders in the world.
Share this article
Telecommunications System & Management received 109 citations as per Google Scholar report
