This document is the synthesis of a study to apply Machine Learning to Argo floats temperature and salinity Delayed Mode Quality Control (Core-Argo-DMQC). There already exist numerous DMQC tests. Most of the times they are specific to a particular type of problem with an Argo profile or an Argo measure. For example, there are tests to detect drifts, other tests to detect spikes, others for thermal lags and so on … To get a clean database, all the alerts generated by those tests sum up and analysts need to study the corresponding profiles.

The aim of our study is to try to use Machine Learning to detect any kind of problem with Argo profiles and reduce the amount of time and work for the analysts.

The model can be upgraded in many ways but it already gets better performances than the existing solution we used to benchmark our model.  For the same detection rate of BAD profiles, the model generate about 25% of alerts less tha the benchmark solution. This document reference the process we developed to get those results.