The increase of connected devices and the constantly evolving methods and techniques by attackers pose a challenge for network intrusion detection systems from conception to operation. As a result, we see a constant adoption of machine learning algorithms for network intrusion detection systems. However, the dataset used by these studies has become obsolete regarding both background and attack traffic. This work describes the AB-TRAP framework that enables the use of updated network traffic and considers operational concerns to enable the complete deployment of the solution. AB-TRAP is a five-step framework consisting of ( i) the generation of the attack dataset, (ii) the bonafide dataset, (iii) training of machine learning models, (iv) realization (implementation) of the models, and ( v) the performance evaluation of the realized model after deployment. We exercised the AB-TRAP for local (LAN) and global (internet) environments to detect TCP port scanning attacks. The LAN study case presented an f1-score of 0.96, and an area under the ROC curve of 0.99 using a decision tree with minimal CPU and RAM usage on kernel-space. For the internet case with eight machine learning algorithms with an average f1-score of 0.95, an average area under the ROC curve of 0.98, and an average overhead of 1.4% CPU and 3.6% RAM on user-space in a single-board computer. This framework has the following paramount characteristics: it is reproducible, uses the most up-to-date network traffic, attacks, and addresses the concerns to the model's realization and deployment.

Wildfires pose a significant threat to ecosystems, human lives, and infrastructure worldwide. Traditional wildfire detection and risk assessment methods often suffer from limitations such as delayed detection and low confidence in certain regions. In this paper, we propose a novel computational system based on machine learning for wildfire risk assessment using data collected by drones. The system can integrate various sensors to capture spatiotemporal data on environmental factors such as temperature, humidity, and vegetation. By leveraging high-resolution data collected through autonomous drone missions, our system enhances wildfire risk estimation and enables proactive mission planning. Although the system is mainly designed to address wildfire monitoring using drone-collected data, it can be easily adapted to other environmental monitoring applications and other sources of data. We demonstrate the effectiveness of our approach through a comprehensive evaluation and validation process in both simulated and real-world environments. Our work contributes to advancing wildfire monitoring capabilities, improving early detection, and mitigating the impact of wildfires on communities and the environment.

This paper addresses an investigation with machine learning (ML) classification techniques to assist in the problem of flash flood now casting. We have been attempting to build a Wireless Sensor Network (WSN) to collect measurements from a river located in an urban area. The machine learning classification methods were investigated with the aim of allowing flash flood now casting, which in turn allows the WSN to give alerts to the local population. We have evaluated several types of ML taking account of the different now casting stages (i.e. Number of future time steps to forecast). We have also evaluated different data representation to be used as input of the ML techniques. The results show that different data representation can lead to results significantly better for different stages of now casting.

Drones can play a game-changing role in reducing both cost and time in the context of last-mile deliveries. This paper addresses the last-mile delivery problem from a complex system viewpoint, where the collective performance of the drones is investigated. We consider a last-mile delivery system with a tradable permit model (TPM) for airspace use. Typically, in other research works regarding lastmile delivery drones, a fully cooperative centralized scenario is contemplated. In our approach, due to the TPM, the agents (i.e. drones) need to compete for airspace permits in a distributed manner. We simulate the system and evaluate how different parameters, such as the arrival rate and airspace dimensions, impact the system behavior in terms of the cost and time needed by the drones to acquire flight permits, and the airspace utilization. We use a simplified simulation model, where the agents’ strategies are naïve, and the drones’ flight dynamics are not accounted for. Nevertheless, the simulation’s level of detail is adequate for capturing interesting properties from the agents’ collective behavior, as our results support. The obtained results show that the system’s performance is satisfactory, even with naïve agents and under high traffic conditions. Moreover, a real-world implementation of our competitive decentralized approach would lead to advantages, such as fast permit transactions, simple computational infrastructures, and error resilience.

The emergence of collective dynamics in neural networks is a mechanism of the animal and human brain for information processing. In this paper, we develop a computational technique using distributed processing elements in a complex network, which are called particles, to solve semisupervised learning problems. Three actions govern the particles' dynamics: generation, walking, and absorption. Labeled vertices generate new particles that compete against rival particles for edge domination. Active particles randomly walk in the network until they are absorbed by either a rival vertex or an edge currently dominated by rival particles. The result from the model evolution consists of sets of edges arranged by the label dominance. Each set tends to form a connected subnetwork to represent a data class. Although the intrinsic dynamics of the model is a stochastic one, we prove that there exists a deterministic version with largely reduced computational complexity; specifically, with linear growth. Furthermore, the edge domination process corresponds to an unfolding map in such way that edges “stretch” and “shrink” according to the vertex-edge dynamics. Consequently, the unfolding effect summarizes the relevant relationships between vertices and the uncovered data classes. The proposed model captures important details of connectivity patterns over the vertex-edge dynamics evolution, in contrast to the previous approaches, which focused on only vertex or only edge dynamics. Computer simulations reveal that the new model can identify nonlinear features in both real and artificial data, including boundaries between distinct classes and overlapping structures of data.