Time Series Segmentation Based on Stationarity Analysis to Improve New Samples Prediction.

A wide range of applications based on sequential data, named time series, have become increasingly popular in recent years, mainly those based on the Internet of Things (IoT). Several different machine learning algorithms exploit the patterns extracted from sequential data to support multiple tasks. However, this data can suffer from unreliable readings that can lead to low accuracy models due to the low-quality training sets available. Detecting the change point between high representative segments is an important ally to find and thread biased subsequences. By constructing a framework based on the Augmented Dickey-Fuller (ADF) test for data stationarity, two proposals to automatically segment subsequences in a time series were developed. The former proposal, called Change Detector segmentation, relies on change detection methods of data stream mining. The latter, called ADF-based segmentation, is constructed on a new change detector derived from the ADF test only. Experiments over real-file IoT databases and benchmarks showed the improvement provided by our proposals for prediction tasks with traditional Autoregressive integrated moving average (ARIMA) and Deep Learning (Long short-term memory and Temporal Convolutional Networks) methods. Results obtained by the Long short-term memory predictive model reduced the relative prediction error from 1 to 0.67, compared to time series without segmentation.

IoTDS: A One-Class Classification Approach to Detect Botnets in Internet of Things Devices.

Internet of Things (IoT) devices have become increasingly widespread. Despite their potential of improving multiple application domains, these devices have poor security, which can be explored by attackers to build large-scale botnets. In this work, we propose a host-based approach to detect botnets in IoT devices, named IoTDS (Internet of Things Detection System). It relies on one-class classifiers, which model only the legitimate device behaviour for further detection of deviations, avoiding the manual labelling process. The proposed solution is underpinned by a novel agent-manager architecture based on HTTPS, which prevents the IoT device from being overloaded by the training activities. To analyse the device's behaviour, the approach extracts features from the device's CPU utilisation and temperature, memory consumption, and number of running tasks, meaning that it does not make use of network traffic data. To test our approach, we used an experimental IoT setup containing a device compromised by bot malware. Multiple scenarios were made, including three different IoT device profiles and seven botnets. Four one-class algorithms (Elliptic Envelope, Isolation Forest, Local Outlier Factor, and One-class Support Vector Machine) were evaluated. The results show the proposed system has a good predictive performance for different botnets, achieving a mean F1-score of 94% for the best performing algorithm, the Local Outlier Factor. The system also presented a low impact on the device's energy consumption, and CPU and memory utilisation.