Boosting Predictions by Calibration of Traffic Model and Learning of Indicators´ Distributions

In this paper, we discuss a two-step scheme using measurements that are taken on a real network. The two steps are complementary and aim to enhance the precision and the quality of a radio network planning tool. We have, in a first step, calibrated the tool by means of live traffic data and/or measurements that are taken on the air interface of a real network and are processed to calculate the traffic values on each cell. We show that the availability of real data is highly valuable since it provides a more detailed view of the network behavior and performance. In a second step, we have proposed a novel algorithm based on a fuzzy Bayesian framework to ameliorate the generalization of a distribution learning system. The learning system aims to correct the predictions of the planning tool and uses the information contained in the simulations as well as the knowledge of the measurements to learn a relation function. The fuzzy Bayesian clustering algorithm is a preprocessing technique that divides the whole learning space into subspaces, where the capacity of the learning system to predict unobserved configurations (generalization) is better performed.