For 6 weeks (from the 17th February to the 2nd April 2020), a technological collaboration project took place involving the Cognitive team from Altice Labs and the Data Science Elite team from IBM. In this project, a reusable machine learning pipeline was developed jointly, which is capable of predicting unavailability events (Site Down) of sites on the MEO mobile network, with a minimum of 6 hours in advance.

The ability to successfully carry out these types of predictions will allow a service provider’s operations center teams to incorporate new work processes that will allow them to implement a preventive approach, abandoning the current essentially reactive logic of responding to operator network failures, fundamentally based on static and deterministic business rules. Additionally, an impact on the experience of MEO mobile network customers will be expected, by minimizing the impact of website failures and by implementing proactive deterrence in call centers.

The challenge we set out to face was quite difficult, in particular due to the sparse, diverse and unbalanced nature of the available alarm data, as well as the complexity of the rules and business data underlying the forecast of this type of failures.

The project allowed the Cognitive team of Altice Labs to develop their skills in Data Science, as well as the incorporation of new knowledge and good practices in their work processes.

The joint work took place in a very intense way, in an environment of great sharing, mutual help and involvement, even with the disturbances resulting from the Covid-19 pandemic, which forced the final 2 weeks to take place in a remote work regime.

In addition to the Cognitive team, the project also involved the Alarm Manager product team, the Ux team (all of these teams from Altice Labs), and had the close collaboration of several teams from Altice/MEO.