Research

Securing cloud microservices requires a unified understanding of how services behave, authenticate, and interact in real time. Unlike existing methods that analyze telemetry signals in isolation, this work presents a heterogeneous graph-based Zero-Trust framework that represents microservices using multi-modal telemetry—logs, metrics, traces, and authentication flows—embedded directly into graph nodes and edges. A Graph Neural Network architecture with attention captures risk propagation across service dependencies, while a joint anomaly detection and trust computation mechanism generates dynamic trust scores with temporal decay to support continuous verification. These trust signals drive real-time dynamic policy enforcement capable of denying or restricting suspicious interactions with minimal operational overhead. Experiments on the TrainTicket, Sock Shop, and DeathStarBench benchmarks show strong performance, achieving 97.2% accuracy, 98.1% recall, and 0.987 AUC on TrainTicket, with consistent results across the other datasets and latency overhead below 3.2 ms. Robustness tests demonstrate accuracy above 95.8% under noisy logs, delayed traces, and authentication failures. Ablation and SHAP analyses confirm that leveraging multiple telemetry modalities—especially authentication data—is critical for accurate detection and trust scoring. These findings show that multi-modal heterogeneous graph modeling, coupled with integrated anomaly-to-policy decision pipelines, provides an effective foundation for Zero-Trust security in cloud-native microservices.

Activity recognition and transportation mode detection are the key research areas for context-aware systems. In smart environments such as cities, buildings, transportation systems etc., ambient intelligence applications watch on human activities in order to increase the quality of transportation, health, traffic and human-oriented services. Literature review shows that existing studies typically deal with recognition of motorized and non-motorized daily activities under the name of human activity recognition and transport mode detection. Only few studies worked on these problems with rich datasets in terms of the number of classes. To the best of our knowledge, for the first time, a study examines nearly the whole motorized/non-motorized transportation modes of people including still, walk, run, climbing upstairs, climbing downstairs, bicycle, motorbike, car, metro, train, high speed rail (HSR), tram and metrobus, excluding ferry and airplane. To outperform existing solutions on such a large dataset, an extended version of the well-known HTC dataset, especially consisting of similar classes, such as train, metro, tram and high speed rail, we propose a combined solution of an Long Short-Term Memory network and Healing algorithm. In our experiments, we first reveal the limits of traditional machine learning algorithms on such number of classes. In addition to that, apart from other studies exploiting deep learning approach, we then examine the potential input types for LSTM network, including raw sensor data, knowledge-based features and features obtained via Auto Encoder. Experimental results show that our proposed idea, feeding knowledge-based features into frames of LSTM network make a remarkable difference and bring out a robust, orientation-independent and generic solution for these well-known problems including activity recognition and transport mode detection. Besides, we examined the hyper-parameters of our deep learning approach and specified the effective parameter set including window size, number of frames, unit size, dropout rate and batch size. Our test results reveal that we could achieve a success rate of 95.5% and outperform the state-of-the art solutions for 12 different transportation modes with the help of our former algorithm, namely Healing, on a totally journey-independent dataset where instances for train, validation and test datasets are selected from different journeys.

In recent years, the research on activity recognition has gained speed especially with the development of smart phones and wearable devices. Activities could be categorized into two main groups. simple activities such as walking, running and complex activities such as eating, sleeping, brushing teeth. In this survey paper, articles about activity recognition are examined thoroughly. Sensors and devices used in activity recognition, types of daily activities, application areas, data collection process, training methods, classification algorithms and resource consumption are mentioned in details. The state of the art is elaborated and the existing methods are compared to each other. Later, open data sets are mentioned and studies offering innovative solutions using latest approaches such as deep learning methods are introduced. Finally, still open issues on this area are presented and future work has been discussed.

Activity recognition and travel mode detection is important topic with application areas in health, smart environment, transportation planning and traffic management domains. The recognition of human activities has become easier due to the widespread use of smartphones and smart watches. While activity recognition and travel mode detection are treated as separate problems in the literature, evaluating them together will provide solutions to wider application domains. In this paper activity recognition and travel mode detection are treated as single problem. In the scope of thesis 13 activities and travel modes are classified using accelerometer, gyroscope and magnetometer sensors that are built in smartphones. Classified activities and travel modes are still, walk, run, upstairs, downstairs, bicycle, motor, car, metro, train, HSR, tram and metrobus. In this thesis, LSTM (Long Short-Term Memory) deep learning method is trained with raw data, with 27 statistical time domain features and with features extracted by Auto Encoders. Using 16 seconds windows and LSTM, we have obtained 86.4% accuracy with time domain features, 83.3% accuracy with raw data, 79.3% accuracy with features obtained from Auto Encoder. Those results are compared with results obtained from Naïve Bayes, Decision Tree, Support Vector Machine and Random Forest machine learning methods. Accuracy results obtained from machine learning methods are %56.9, %71.2, %81.6 and %83.8 in order. 93.1% accuracy is obtained after healing of best result obtained using LSTM.