Artificial Neural Networks
Neural-network intrusion detection on the DARPA/MIT Lincoln dataset, plus NDVI-based crop prediction with ANNs.
Neural Networks Detecting Intrusions And Attacks On Computer Networks
Intrusion Detection Systems (IDS) are software or hardware systems that automate the process of monitoring the events occurring in a computer system or network, analysing them for signs of security problems. They are an especially important component of computer network security architecture given the increased exposure of business-critical process, product, financial, customer and employee internal systems and data to the internet, including via cloud deployment, IoT, mobile self-service applications, B2B and B2C integration.
With network firewalls, inbound network traffic is filtered according to a set of predefined rules, meaning malicious traffic can reach systems if no rules exist that match and flag such traffic. Signature-based detection systems similarly rely on an inventory of signature definitions to enable flagging of activity as a threat. In both cases the technique is reactive: the firewall rules and signature inventory are updated only after an attack type is known.
Conversely, IDS is both proactive and able to deal with unknown threats and abnormal activity, providing an additional security layer by actively classifying internal and external traffic in real time as good or bad according to characteristics of the network data packets, and can respond appropriately by closing ports, blacklisting IP addresses, or sending TCP reset packets to break connections.
This study implements Neural Networks as the basis of an IDS that classifies computer network activity as bad (covering intrusions and attacks) and good (normal connections). A dataset originally prepared by MIT Lincoln Labs for the 1998 DARPA Intrusion Detection Evaluation Program, simulating air force base LAN activity, is used.
Artifacts
- Neural network model detecting intrusions or attacks on a computer network - grade 80%
- Source code on GitHub
Neural Networks For Crop Prediction And Monitoring Ecosystems
With the environment under increasing pressure from climate change and human development, there appear to be real opportunities to benefit from smarter farming ecosystem management given today's perfect technology storm of available time-series satellite data, accessible machine learning software, elastic and affordable cloud computing, connected IoT and more capable UAV platforms.
Knowing exactly when to harvest for maximum crop yield, and planning for it across the complete supply chain, brings significant benefit. So does knowing when to selectively apply herbicides, or take action on newly identified deforestation or areas at risk of landslide.
This study reviews recent application of artificial neural networks for crop prediction, classification and monitoring of ecosystems. It focuses on the use of NDVI (Normalised Difference Vegetation Index) data, which measures the difference between near-infrared (which vegetation strongly reflects) and red light (which vegetation absorbs), sourced from satellites and UAVs.
Artifacts