Geolocalized Transductive Graph-Based Regression Applied to Sustainability Indicators Prediction

Sustainability Indicators are gaining notoriety and importance in the world. However, restrictive reasons for accessing information or locations generate the demand for predicting such indicators. Here, we present a Transductive Graph-based Regression framework for this task. The framework generates a hybrid graph in two steps: the first step generates the connections among vertices, with topological variations: Regular, Small-World, and Scale-Free. The second step adopts the Spectral topology yielding to the weights associated with the edges. For the inference, graph-based techniques, such as random walk and linear regression, combine propagation of the indicators and an evaluation process based on Gibbs sampling. An empirical analysis of different datasets shows that our strategy surpasses traditional approaches considering the measures used in the evaluation. The experiments show that when employing information from the dataset or geolocation data during the graph construction. Comparing the proposed framework with state-of-the-art baselines proves that our solutions outperform the previous techniques.