Deep Learning Framework for Proactive Network Fault Diagnosis in Cloud Infrastructure Arpit U. Chaudhari, Sumedh P. Ingale, Anand A. Chaudhari, Ankit R. Mune, Swapnil Anil Kale, Aditya O. Sable 2025 3rd Dmiher International Conference on Artificial Intelligence in Healthcare Education and Industry Idicaihei 2025, 2025 A lot of mission-critical apps run in the cloud, and for those to keep running smoothly, they need networking that is stable and can handle problems. Traditional rule-based and threshold-driven methods of finding network faults don't always work with the changing and nonlinear nature of large-scale cloud systems. This means that faults are found later and there is more downtime. To get around these problems, this paper suggests using deep learning to find network problems before they happen in cloud systems. The system uses advanced neural designs, like recurrent and convolutional models, to look at performance metrics like delay, packet loss, bandwidth usage, and server logs in real time. By modelling how events depend on each other over time and pulling out high-level feature models, the system can spot possible problems before they become service failures. The proposed approach combines finding strange things and correctly classifying them as faults. This lets network problems be found early and correctly categorized. When tested on standard datasets and virtual cloud settings, the framework does a better job of accuracy, precision, and memory compared to other machine learning methods. This proactive method cuts down on the mean time to discovery (MTTD) and mean time to recovery (MTTR) by a large amount. This makes service uptime better, lowers running costs, and makes sure that users of cloud-based systems have a smooth experience.
Optimization Strategies for Cyber Threat Detection in Cloud Architectures Leveraging Deep Machine Learning for Advanced Malware Identification Ankit R. Mune, Arpit U. Chaudhari, M. A. Pund, Tejas P. Adhau, Sumedh P. Ingale, Aditya O. Sable 2024 2nd Dmiher International Conference on Artificial Intelligence in Healthcare Education and Industry Idicaiei 2024, 2024 The fast growth of cloud designs has created new security holes, which makes finding cyber threats a very important issue. This study looks into improvement methods that are meant to make it easier to find cyber threats in cloud settings. Using deep machine learning methods, the study aims to make it easier to spot advanced malware, which is very dangerous for cloud infrastructure. The recommended strategy combines optimization strategies with profound learning models to move forward the speed and precision of recognizable proof. The strategy incorporates looking at huge sets of information from cloud frameworks to educate models that can discover complex malware patterns. To discover out how well the moved forward discovery framework works, execution measures just like the rate of location, the rate of untrue positives, and the taken a toll of computing are looked at. The study comes about appear that profound machine learning is much superior at finding malware than standard strategies. This appears that profound machine learning has the capacity to create cloud security more grounded. This consider makes a difference make solid, versatile ways to discover cyber dangers in cloud frameworks. It gives us useful information for making security way better in a world that's becoming increasingly advanced.
An analysis of heterogeneous data with extreme learning via unsupervised multiple kernels Ankit R. Mune, Sohel A. Bhura 2nd International Conference on Data Engineering and Applications Idea 2020, 2020 Multiple sources without information about labels collect a large amount of data, often containing heterogeneous data, namely various types, structures as well as distributions. Such data can include Instagram, Twitter, and Facebook and YouTube, texts, images and videos. Advanced unsupervised learning methods (with multiple kernels) cabs are applied to extract information from such great unlabeled heterogeneous data MKL Traditional MKL algorithms are a good way to reveal information from multiple sources. Some efforts in managing difficult, heterogeneous distributed data, for example using the kernel, have currently been made to efficiently capture data from heterogeneous data. In recent times some unsupervised learning efforts have been made. In this paper, we illustrate the problem by heterogeneous knowledge and discuss various kernel and extreme learning approaches and their problems without supervision.
