гібридна модель

Hybrid Behavioural Analysis Method for Early Detection of Anomalous Activity in Web Applications

The research introduces a hybrid behavioural analysis technique for early detection of anomalous user behavior observed on web applications. This strategy involves statistical probability modeling and sequence- based deep learning to design interpretable and robust anomaly detection. A probability baseline has been obtained as a probabilistic basis using KDE (Kernel Density Estimation) and longitudinal time series patterns are sampled using an LSTM network. The hybrid anomaly score combines these two models to dynamically analyze behavioural deviations.

MACHINE LEARNING-BASED PREDICTION OF ELECTRIC VEHICLE REMAINING RANGE WITH CONSIDERATION OF BATTERY DEGRADATION

Accurate prediction of the remaining driving range in electric vehicles (EVs) is critical for efficient trip planning, reducing the risk of battery depletion, and improving user experience. One of the significant challenges in achieving high prediction accuracy is battery degradation, which gradually reduces battery capacity and impacts the vehicle’s range. This study uses machine learning algorithms to investigate the impact of incorporating battery degradation—expressed through the State of Health (SoH) indicator—into range prediction models.