Bike-sharing is adopted as a valid alternative to traditional public transports since they are ecofriendly, prevent traffic congestions, reduce the probability of social contacts which happens on most of the public means. However, some problems may occur such as the irregular distribution of bikes on the stations/racks/areas, and the difficulty of knowing in advance the rack status with a certain degree of confidence, whether there will be available bikes at a specific bike-station at a certain time of the day, or a free slot for leaving the rented bike. Thus, providing predictions can be useful to improve the quality of service, especially in those cases in which the bike racks are used for e-bikes, which need to be recharged. This paper compares the state-of-the-art techniques to predict the number of available bikes and free bikeslots in bike-sharing stations (i.e., bike racks). To this end, a set of features and predictive models were compared to identify the best models and predictors for short-term predictions of 15, 30, 45, and 60 minutes. The study demonstrated that deep learning and in particular Bidirectional Long Short-Term Memory networks (Bi-LSTM) offer a robust approach for the implementation of reliable and fast predictions of available bikes, even with a limited amount of historical data. The paper also reported an analysis of feature relevance based on SHAP that demonstrated the validity of the model for different behaviours of the clusters. The solution and its validation were derived by using data collected in bike-stations in the cities of Siena and Pisa (Italy), in the context of Sii-Mobility National Research Project on Mobility and Transport and Snap4City Smart City IoT infrastructure.

Rainfall induced landslide is one of the main geological hazard in Italy and in the world. Each year it causes fatalities, casualties and economic and social losses on large populated areas. Accurate shortterm predictions of landslides can be extremely important and useful, in order to both provide local authorities with efficient prediction/early warning and increase the resilience to manage emergencies. There is an extensive literature addressing the problem of computing landslide susceptibility maps (which is a classification problem exploiting a large range of static features) and only few on actual short terms predictions (spatial and temporal). The short-term prediction models are still empirical and obtain unsatisfactory results, also in the identification of the predictors. The new aspects addressed in this paper are: (i) a short-term prediction model (1 day in advance) of landslide based on machine learning, (ii) real time features as good predictors. The introduction of explainable artificial intelligence techniques allowed to understand global and local feature relevance. In order to find the best prediction model, a number of machine learning solutions have been implemented and assessed. The models obtained overcome those of the literature. The validation has been performed in the context of the Metropolitan City of Florence, data from 2013 to 2019. The method based on XGBoost achieved best results, demonstrating that it is the most reliable and robust against false alarms. Finally, we applied explainable artificial intelligence techniques locally and globally to derive a deep understand of the predictive model's outputs and features' relevance, and relationships. The analysis allowed us to identify the best feature for short term predictions and their impact in the local cases and global prediction model. Solutions have been implemented on Snap4City.org infrastructure.

Nowadays, traffic management and sustainable mobility are central topics for intelligent transportation systems (ITS). Thanks to new technologies, it is possible to collect real-time data to monitor the traffic situation and contextual information by sensors. An important challenge in ITS is the ability to predict road traffic flow data. The short-term predictions (10-60 minutes) of traffic flow data is a complex nonlinear task that has been the subject of many research efforts in past few decades. Accessing traffic flow data is mandatory for a large number of applications that have to guarantee a high level of services such as traffic flow analysis, traffic flow reconstruction, which in their turn are used to compute predictions needed to perform what-if analysis, forecast routing, conditioned routing, predictions of pollutant, etc. This paper proposes a solution for short-term prediction of traffic flow data by using a architecture capable to exploit Convolutional Bidirectional Deep Long Short Term Memory neural networks (CONV-BI-LSTM). The solution adopts a different architecture and features, so as to overcome the state-of-the-art solutions and provides precise predictions addressing traffic flow data in cities, which are tendentially very noisy with respect to the ones measured in high-speed roads, the latter being the validation context for the majority of state-of-the-art solutions. The proposed solution has been developed and validated in the city context and data via Sii-Mobility, a smart city mobility and transport national project and it is currently in use in other contexts such as in Snap4City PCP EC, TRAFAIR CEF, and REPLICATE H2020 SCC1, and it is operative in those areas.

No abstract