Optimizing the planning of FTTH deployments through predictive Machine Learning models..

Fiber-to-the-Home (FTTH) is a fundamental technology for the digital transformation and economic development of our societies. Its high speed, capacity, and reliability enable an enhanced user experience and facilitate a broad range of digital services. Moreover, it contributes to reducing the digital divide and improving social cohesion. However, the substantial investment in deploying this infrastructure poses a significant challenge for telecommunications operators. Therefore, proper planning of these activities is critical.

In this context, berocam was hired by a telecommunications operator in a European country with highly ambitious deployment goals for fiber across the entire national territory. This encompassed urban areas, where the operator had more experience, as well as rural areas with low population density.

The project’s goal was precisely to provide this operator with tools for conducting a planning and budgeting process for the FTTH deployment in a much more agile, precise, and detailed manner than was possible up to that point.

To achieve the objective, we proposed an analytical approach based on Machine Learning that leveraged both the client’s internal historical information regarding past deployments and external sources, including data related to municipalities and deployment areas. This involved considering sociodemographic, geographic, topological, and urbanistic factors. With all this, the challenge was to harness that large volume of data through the construction of a predictive model capable of inferring the cost of future deployments.

One of the main focuses of the project was calculating the distance in meters of a route that could be likened to the actual path of the fiber cables during deployment in a municipality. In other words, a layout that could connect each building to the nearest communication hub while respecting the urban details of each municipality. We approached this calculation as a graph problem and applied Minimum Spanning Tree (MST) algorithms to obtain the result. These algorithms help find the minimum spanning tree that connects the vertices of a graph.

In our case, the starting point of the graph would be the telecommunication hub, and the vertices would represent the buildings to be connected. After processing all the geolocated information about the central hub and buildings, and incorporating external sources detailing the urban and topological configuration of each project into the algorithm, we were able to carry out the detailed distance calculation for each project.

We integrated this result with dozens of additional variables, both external and internal. We initiated an iterative process where we tested both linear and non-linear approaches, applying cross-validation exercises at all stages of the analysis to ensure the representativeness, robustness, and accuracy of the model.

As the culmination of the process, we managed to generate a multivariable linear regression model with very high levels of correlation (R2 > 0,8) and the prediction results for fiber meters and associated deployment costs turned out to be very accurate when compared to real cases.

Applying this predictive tool based on Machine Learning, our client can now predict, in an automated manner, with an error range of less than +/-5%, the cost of all types of FTTH deployments, regardless of the typology and region. This represents a very significant improvement over the deviations recorded in previous budgeting and planning processes.