Riccardo Taormina

ANN-based interval forecasting of streamflow discharges using the LUBE method and MOFIPS

The estimation of prediction intervals (PIs) is a major issue limiting the use of Artificial Neural Networks (ANN) solutions for operational streamflow forecasting. Recently, a Lower Upper Bound Estimation (LUBE) method has been proposed that outperforms traditional techniques for ANN-based PI estimation. This method construct ANNs with two output neurons that directly approximate the lower and upper bounds of the PIs. The training is performed by minimizing a coverage width-based criterion (CWC), which is a compound, highly nonlinear and discontinuous function. In this work, we test the suitability of the LUBE approach in producing PIs at different confidence levels (CL) for the 6h ahead streamflow discharges of the Susquehanna and Nehalem Rivers, US. Due to the success of Particle Swarm Optimization (PSO) in LUBE applications, variants of this algorithm have been employed for CWC minimization. The results obtained are found to vary substantially depending on the chosen PSO paradigm. While the returned PIs are poor when single-objective swarm optimization is employed, substantial improvements are recorded when a multi-objective framework is considered for ANN development. In particular, the Multi-Objective Fully Informed Particle Swarm (MOFIPS) optimization algorithm is found to return valid PIs for both rivers and for the three CL considered of 90%, 95% and 99%. With average PI widths ranging from a minimum of 7% to a maximum of 15% of the range of the streamflow data in the test datasets, MOFIPS-based LUBE represents a viable option for straightforward design of more reliable interval-based streamflow forecasting models. Lower Upper Bound Estimation (LUBE) used to build streamflow prediction intervals.LUBE method enhanced via Multi-Objective Fully Informed Particle Swarm (MOFIPS).MOFIPS-based LUBE develops reliable ANN streamflow forecasting models.

Characterizing Cyber-Physical Attacks on Water Distribution Systems

AbstractThis work contributes a modeling framework to characterize the effect of cyber-physical attacks (CPAs) on the hydraulic behavior of water distribution systems. The framework consists of an ...

Identifying (Quasi) Equally Informative Subsets in Feature Selection Problems for Classification: A Max-Relevance Min-Redundancy Approach

An emerging trend in feature selection is the development of two-objective algorithms that analyze the tradeoff between the number of features and the classification performance of the model built with these features. Since these two objectives are conflicting, a typical result stands in a set of Pareto-efficient subsets, each having a different cardinality and a corresponding discriminating power. However, this approach overlooks the fact that, for a given cardinality, there can be several subsets with similar information content. The study reported here addresses this problem, and introduces a novel multiobjective feature selection approach conceived to identify: 1) a subset that maximizes the performance of a given classifier and 2) a set of subsets that are quasi equally informative, i.e., have almost same classification performance, to the performance maximizing subset. The approach consists of a wrapper [Wrapper for Quasi Equally Informative Subset Selection (W-QEISS)] built on the formulation of a four-objective optimization problem, which is aimed at maximizing the accuracy of a classifier, minimizing the number of features, and optimizing two entropy-based measures of relevance and redundancy. This allows conducting the search in a larger space, thus enabling the wrapper to generate a large number of Pareto-efficient solutions. The algorithm is compared against the mRMR algorithm, a two-objective wrapper and a computationally efficient filter [Filter for Quasi Equally Informative Subset Selection (F-QEISS)] on 24 University of California, Irvine, (UCI) datasets including both binary and multiclass classification. Experimental results show that W-QEISS has the capability of evolving a rich and diverse set of Pareto-efficient solutions, and that their availability helps in: 1) studying the tradeoff between multiple measures of classification performance and 2) understanding the relative importance of each feature. The quasi equally informative subsets are identified at the cost of a marginal increase in the computational time thanks to the adoption of Borg Multiobjective Evolutionary Algorithm and Extreme Learning Machine as global optimization and learning algorithms, respectively.