Towards performance evaluation prediction in WSNs using artificial neural network multi-perceptron

The use of formal methods in specifying and verifying WSNs protocols attracts several researchers. However, based practically on state-space exploration, these formal methods face scalability problems when the specified system is complicated, which is often the case with WSNs protocols. To overcome this last problem, this paper proposes exploiting a machine learning approach to make predictions when the specified system becomes highly complex due to the increasing number of nodes. The main contribution of this paper is the application of a Multi-Layer Perceptron (MLP) to predict a set of crucial performance metrics of the CSMA/CA MAC protocol based on the historical data generated from formal models representing the whole behaviour of the WSN, including waiting time (WT), delay performance (DP), waiting time for an acknowledgement (WTA), and the throughput (TH). The empirical results demonstrate the effectiveness of the proposed MLP architecture compared to other ML techniques (SVR and LR) using various evaluation criteria (MAE, MSE, RMSE). MLP gives the best, and minimum criteria values on all performance metrics datasets in terms of MSE values are around 4.90, 3.11, 8.88, and 4 x 10(-5) for metrics WT, WTA, DP, and TH, respectively. The obtained results in this paper proved the efficiency of the combination between formal models (i.e. Hierarchical Timed Coloured Petri Nets) and machine learning approaches that use artificial neural networks.