Franco Alberto Cardillo
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Research Subjects
Franco Alberto Cardillo has a total of 21 co-authors. Here are the top co-authors by number of shared publications:
- Umberto Straccia
CNR-ISTI
3 shared publications - Franca Debole
CNR-ISTI Pisa Italy
2 shared publications - Marcello Ferro
Institute for Computational Linguistics “A. Zampolli”, Italian National Research Council, 56124 Pisa, Italy
2 shared publications - Vito Pirrelli
Institute for Computational Linguistics “A. Zampolli”, Italian National Research Council, 56124 Pisa, Italy
2 shared publications Alvise Papa
1 shared publicationAngela Andrigo
1 shared publication
Latest Publications
- Open Access
- 14 March 2026
- Published by Copernicus GmbH
A Preliminary Analysis of High Water Events in Venice Based on Multi-Decadal Observations and Clustering
High water events in Venice are a recurrent phenomenon, as the city is located only slightly above mean sea level and is directly influenced by water-level variations within the lagoon. Flooding occurs when several physical processes act in combination. The astronomical tide determines the baseline water level, which is subsequently modulated by seiche oscillations in the Adriatic Sea, meteorological forcing (e.g. wind stress and atmospheric pressure), and slower, low-frequency geophysical processes and sea level rise. When these factors co-occur, even if individually moderate, large portions of the city may experience flooding.Repeated flooding has significant economic and social impacts, limits pedestrian and naval traffic and contributes to the degradation of buildings and cultural heritage. To mitigate these effects, a range of protective measures is implemented and coordinated by an early warning system. The effectiveness of these measures depends on their timely activation. However, mitigation actions are associated with substantial economic costs and may themselves generate negative impacts if deployed unnecessarily. For instance, interruptions to public transport services affect daily activities, while the operation of the MOSE barrier entails considerable financial costs. Accurate and reliable forecasts are therefore essential to balance flood protection with the economic and social costs of mitigation measures.Current forecasting systems primarily estimate water levels and peak values, and these are typically estimated at a limited number of locations. These systems are based on sophisticated statistical and hydrodynamic models. Although they perform well in most situations, their accuracy can be affected by uncertainties in atmospheric forcing and by limitations in representing the full variability of high water events. This work explores the potential of complementary approaches based on the analysis of observational data rather than explicit physical modelling.Data-driven approaches, in particular Machine Learning (ML) methods, analyze historical data without relying on predefined, human-designed model structures. ML models are able to capture recurring patterns and complex feature interactions that are difficult to incorporate into traditional numerical models. Among these approaches, clustering techniques aim to identify recurrent types of events based on similarities in their temporal evolution and associated meteorological conditions. This enables events characterized by similar water levels to be differentiated according to the combinations of underlying meteorological drivers, thereby providing additional information to support forecasting and response planning.In this work, we present a preliminary analysis based on several clustering approaches, including k-means, DBSCAN, and deep learning–based methods, applied to a multi-decadal atmospheric dataset and to the longest available reconstructed hourly sea-level records for the northern Adriatic Sea, specifically developed for this study. We compare the resulting event classifications and discuss how cluster-derived information may complement existing forecasting systems in support of flood-mitigation strategies for the city of Venice.
- 14 March 2026
- Published by Copernicus GmbH
Open AccessA Preliminary Analysis of High Water Events in Venice Based on Multi-Decadal Observations and Clustering
High water events in Venice are a recurrent phenomenon, as the city is located only slightly above mean sea level and is directly influenced by water-level variations within the lagoon. Flooding occurs when several physical processes act in combination. The astronomical tide determines the baseline water level, which is subsequently modulated by seiche oscillations in the Adriatic Sea, meteorological forcing (e.g. wind stress and atmospheric pressure), and slower, low-frequency geophysical processes and sea level rise. When these factors co-occur, even if individually moderate, large portions of the city may experience flooding.Repeated flooding has significant economic and social impacts, limits pedestrian and naval traffic and contributes to the degradation of buildings and cultural heritage. To mitigate these effects, a range of protective measures is implemented and coordinated by an early warning system. The effectiveness of these measures depends on their timely activation. However, mitigation actions are associated with substantial economic costs and may themselves generate negative impacts if deployed unnecessarily. For instance, interruptions to public transport services affect daily activities, while the operation of the MOSE barrier entails considerable financial costs. Accurate and reliable forecasts are therefore essential to balance flood protection with the economic and social costs of mitigation measures.Current forecasting systems primarily estimate water levels and peak values, and these are typically estimated at a limited number of locations. These systems are based on sophisticated statistical and hydrodynamic models. Although they perform well in most situations, their accuracy can be affected by uncertainties in atmospheric forcing and by limitations in representing the full variability of high water events. This work explores the potential of complementary approaches based on the analysis of observational data rather than explicit physical modelling.Data-driven approaches, in particular Machine Learning (ML) methods, analyze historical data without relying on predefined, human-designed model structures. ML models are able to capture recurring patterns and complex feature interactions that are difficult to incorporate into traditional numerical models. Among these approaches, clustering techniques aim to identify recurrent types of events based on similarities in their temporal evolution and associated meteorological conditions. This enables events characterized by similar water levels to be differentiated according to the combinations of underlying meteorological drivers, thereby providing additional information to support forecasting and response planning.In this work, we present a preliminary analysis based on several clustering approaches, including k-means, DBSCAN, and deep learning–based methods, applied to a multi-decadal atmospheric dataset and to the longest available reconstructed hourly sea-level records for the northern Adriatic Sea, specifically developed for this study. We compare the resulting event classifications and discuss how cluster-derived information may complement existing forecasting systems in support of flood-mitigation strategies for the city of Venice.
- Open Access
- 13 March 2026
- Published by Copernicus GmbH
Missing data imputation in epidemiology: a comparison between MICE and Machine Learning methods
Missing data represents a challenge in large-scale epidemiological studies as it can introduce a strong and negative bias in the final estimates when not handled appropriately. This issue is particularly relevant in environment health research due to complex relationships between the exposure to risk factors and delayed outcomes. In this work, we evaluate the effectiveness of statistical and Machine Learning (ML) approaches to fill in missing values in data we collected to assess the potential impact on public health of gold mining activities in the Ecuadorian Amazon.There is growing concern regarding the adverse effects on human health in the Ecuadorian Amazon caused by the environmental impact of gold mining activities in the area. To investigate potential associations with adverse birth outcomes, we collected data published by the Ecuadorian National Institute of Statistics and Census (INEC) relative to the annual live birth and fetal death cases in the years from 2014 to 2023. As it is typical in large-scale epidemiological studies, the data contain a proportion of missing values, likely related to the registration and the data entry process. Addressing missing values is considered important for the correct assignment of cases from one hand and the characterisation of risk factors from another. Furthermore, it enables the modelling process when searching for associations between exposure and outcome without erroneous under- or over-reporting of odds ratios (Type I and Type II errors). Currently, the most common approach in epidemiology is to use statistical methods and, specifically, Multivariate Imputation by Chained Equations (MICE), normally instantiated with parametric conditional models. MICE imputes missing values by repeatedly predicting each incomplete variable from the others using standard regression models. In most applications, these predictions rely on linear or generalised linear relationships between variables. This can reduce its effectiveness in predicting missing values in presence of complex, non-linear interactions about variables. Machine Learning represents an interesting alternative as it capture complex, non-linear relationships beyond the linear models typically assumed in MICE, are more flexible with respect to departures from missing-at-random patterns, and reduce the risk of model misspecification by relying on data-driven, implicit model selection rather than requiring the analyst to pre-specify an imputation model.In this study, we present a robust experimental comparison between MICE and several ML-based imputation approaches applied to the Ecuadorian birth data. We assess their performance and discuss the respective strengths and limitations within an epidemiological context.
- 13 March 2026
- Published by Copernicus GmbH
Open AccessMissing data imputation in epidemiology: a comparison between MICE and Machine Learning methods
Missing data represents a challenge in large-scale epidemiological studies as it can introduce a strong and negative bias in the final estimates when not handled appropriately. This issue is particularly relevant in environment health research due to complex relationships between the exposure to risk factors and delayed outcomes. In this work, we evaluate the effectiveness of statistical and Machine Learning (ML) approaches to fill in missing values in data we collected to assess the potential impact on public health of gold mining activities in the Ecuadorian Amazon.There is growing concern regarding the adverse effects on human health in the Ecuadorian Amazon caused by the environmental impact of gold mining activities in the area. To investigate potential associations with adverse birth outcomes, we collected data published by the Ecuadorian National Institute of Statistics and Census (INEC) relative to the annual live birth and fetal death cases in the years from 2014 to 2023. As it is typical in large-scale epidemiological studies, the data contain a proportion of missing values, likely related to the registration and the data entry process. Addressing missing values is considered important for the correct assignment of cases from one hand and the characterisation of risk factors from another. Furthermore, it enables the modelling process when searching for associations between exposure and outcome without erroneous under- or over-reporting of odds ratios (Type I and Type II errors). Currently, the most common approach in epidemiology is to use statistical methods and, specifically, Multivariate Imputation by Chained Equations (MICE), normally instantiated with parametric conditional models. MICE imputes missing values by repeatedly predicting each incomplete variable from the others using standard regression models. In most applications, these predictions rely on linear or generalised linear relationships between variables. This can reduce its effectiveness in predicting missing values in presence of complex, non-linear interactions about variables. Machine Learning represents an interesting alternative as it capture complex, non-linear relationships beyond the linear models typically assumed in MICE, are more flexible with respect to departures from missing-at-random patterns, and reduce the risk of model misspecification by relying on data-driven, implicit model selection rather than requiring the analyst to pre-specify an imputation model.In this study, we present a robust experimental comparison between MICE and several ML-based imputation approaches applied to the Ecuadorian birth data. We assess their performance and discuss the respective strengths and limitations within an epidemiological context.
- 12 June 2024
- Published by Elsevier in Fuzzy Sets and Systems
- Vol. 490
PN-OWL: A two-stage algorithm to learn fuzzy concept inclusions from OWL 2 ontologies
- 12 June 2024
- Published by Elsevier in Fuzzy Sets and Systems
- Vol. 490
PN-OWL: A two-stage algorithm to learn fuzzy concept inclusions from OWL 2 ontologies
- 5 April 2023
- Published by Springer Nature
- p. 58-74
Baselines for Automatic Medical Image Reporting
- 1 June 2022
- Published by Elsevier in Fuzzy Sets and Systems
- Vol. 438, 164-186
Fuzzy OWL-Boost: Learning fuzzy concept inclusions via real-valued boosting
- 1 June 2022
- Published by Elsevier in Fuzzy Sets and Systems
- Vol. 438, 164-186
Fuzzy OWL-Boost: Learning fuzzy concept inclusions via real-valued boosting
- Open Access
Supermetric Search
Metric search is concerned with the efficient evaluation of queries in metric spaces. In general,a large space of objects is arranged in such a way that, when a further object is presented as a query, those objects most similar to the query can be efficiently found. Most mechanisms rely upon the triangle inequality property of the metric governing the space. The triangle inequality property is equivalent to a finite embedding property, which states that any three points of the space can be isometrically embedded in two-dimensional Euclidean space. In this paper, we examine a class of semimetric space which is finitely four-embeddable in three-dimensional Euclidean space. In mathematics this property has been extensively studied and is generally known as the four-point property. All spaces with the four-point property are metric spaces, but they also have some stronger geometric guarantees. We coin the term supermetric space as, in terms of metric search, they are significantly more tractable. Supermetric spaces include all those governed by Euclidean, Cosine, Jensen-Shannon and Triangular distances, and are thus commonly used within many domains. In previous work we have given a generic mathematical basis for the supermetric property and shown how it can improve indexing performance for a given exact search structure. Here we present a full investigation into its use within a variety of different hyperplane partition indexing structures, and go on to show some more of its flexibility by examining a search structure whose partition and exclusion conditions are tailored, at each node, to suit the individual reference points and data set present there. Among the results given, we show a new best performance for exact search using a well-known benchmark.
Open AccessSupermetric Search
Metric search is concerned with the efficient evaluation of queries in metric spaces. In general,a large space of objects is arranged in such a way that, when a further object is presented as a query, those objects most similar to the query can be efficiently found. Most mechanisms rely upon the triangle inequality property of the metric governing the space. The triangle inequality property is equivalent to a finite embedding property, which states that any three points of the space can be isometrically embedded in two-dimensional Euclidean space. In this paper, we examine a class of semimetric space which is finitely four-embeddable in three-dimensional Euclidean space. In mathematics this property has been extensively studied and is generally known as the four-point property. All spaces with the four-point property are metric spaces, but they also have some stronger geometric guarantees. We coin the term supermetric space as, in terms of metric search, they are significantly more tractable. Supermetric spaces include all those governed by Euclidean, Cosine, Jensen-Shannon and Triangular distances, and are thus commonly used within many domains. In previous work we have given a generic mathematical basis for the supermetric property and shown how it can improve indexing performance for a given exact search structure. Here we present a full investigation into its use within a variety of different hyperplane partition indexing structures, and go on to show some more of its flexibility by examining a search structure whose partition and exclusion conditions are tailored, at each node, to suit the individual reference points and data set present there. Among the results given, we show a new best performance for exact search using a well-known benchmark.
- Open Access
- 1 January 2017
- Published by OpenEdition
- p. 77-82
How “deep” is learning word inflection?
Machine learning offers two basic strategies for morphology induction: lexical segmentation and surface word relation. The first one assumes that words can be segmented into morphemes. Inducing a novel inflected form requires identification of morphemic constituents and a strategy for their recombination. The second approach dispenses with segmentation: lexical representations form part of a network of associatively related inflected forms. Production of a novel form consists in filling in one empty node in the network. Here, we present the results of a recurrent LSTM network that learns to fill in paradigm cells of incomplete verb paradigms. Although the process is not based on morpheme segmentation, the model shows sensitivity to stem selection and stem-ending boundaries.
- 1 January 2017
- Published by OpenEdition
- p. 77-82
Open AccessHow “deep” is learning word inflection?
Machine learning offers two basic strategies for morphology induction: lexical segmentation and surface word relation. The first one assumes that words can be segmented into morphemes. Inducing a novel inflected form requires identification of morphemic constituents and a strategy for their recombination. The second approach dispenses with segmentation: lexical representations form part of a network of associatively related inflected forms. Production of a novel form consists in filling in one empty node in the network. Here, we present the results of a recurrent LSTM network that learns to fill in paradigm cells of incomplete verb paradigms. Although the process is not based on morpheme segmentation, the model shows sensitivity to stem selection and stem-ending boundaries.
- Open Access
- 1 January 2016
- Published by OpenEdition
- p. 146-151
Written word production and lexical self-organisation: evidence from English (pseudo)compounds
Elevation in typing latency for the initial letter of the second constituent of an English compound, relative to the latency for the final letter of the first constituent of the same compound, provides evidence that implementation of a motor plan for written compound production involves smaller constituents, in both semantically transparent and semantically opaque compounds. We investigate here the implications of this evidence for algorithmic models of lexical organisation, to show that effects of differential perception of the internal structure of compounds and pseudo-compounds can also be simulated as peripheral stages of lexical access by a self-organising connectionist architecture, even in the absence of morphosemantic information. This complementary evidence supports a maximization-of-opportunity approach to lexical modelling, accounting for the integration of effects of pre-lexical and lexical access.
- 1 January 2016
- Published by OpenEdition
- p. 146-151
Open AccessWritten word production and lexical self-organisation: evidence from English (pseudo)compounds
Elevation in typing latency for the initial letter of the second constituent of an English compound, relative to the latency for the final letter of the first constituent of the same compound, provides evidence that implementation of a motor plan for written compound production involves smaller constituents, in both semantically transparent and semantically opaque compounds. We investigate here the implications of this evidence for algorithmic models of lexical organisation, to show that effects of differential perception of the internal structure of compounds and pseudo-compounds can also be simulated as peripheral stages of lexical access by a self-organising connectionist architecture, even in the absence of morphosemantic information. This complementary evidence supports a maximization-of-opportunity approach to lexical modelling, accounting for the integration of effects of pre-lexical and lexical access.