Amjan Shaik

Amjan Shaik

Auto-generated profile, based on publicly available information

Amjan Shaik

St.Peter’s Engineering College

Research Focus & Metrics

Academic Career Length

5Years (2022-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
30 December 2024 in International Journal of Computational and Experimental Science and Engineering
Verified
Journal Article
Open access
PDF file
ASAmjan Shaik
Amjan Shaik
BUBhuvan Unhelkar
Bhuvan Unhelkar
PCPrasun Chakrabarti
Prasun Chakrabarti
The fast growth of IO networks has resulted in a security crisis besides the development of decentralized-based innovations, and such decentralized bases or technologies also made challenges in terms of speed, performance, and scalability. Traditional machine learning-based intrusion detection systems (IDS) are unable to manage the intricate and non-linear correlations seen in massive amounts of IoT data. They produce relatively low detection rates, especially in multi-class classification, where many attack types must be addressed. Overcoming these hurdles calls for frameworks: innovative enough to accommodate the challenge whilst using the wealth of data produced by IoT devices. Abstract In this paper, we introduce a unique MLP-based deep learning architecture for intrusion detection in IoT settings. This framework includes a preprocessing pipeline that optimally normalizes and applies one-hot-encoding to the data to prepare it optimally for classification. We tested the algorithms on the UNSW-NB15 dataset, commonly used for IDS. Mere quantitative results show that MLP surpasses classical models like Logistic Regression, SVM, and Random Forests, giving a precision of 97.53%, recall of 97.23%, and accuracy of 97.73% on the multi-class classification task. This framework is undoubtedly scalable and provides a sufficient security mechanism for the whole IoT ecosystem; hence, it can be used in various actual use cases. This performance shows that it could solve the new threats developing in IoT environments.

1 June 2023
Verified
Conference Paper
Closed access
PVp vidyullatha
p vidyullatha
VRV. Dinesh Reddy
V. Dinesh Reddy
KRK. Dasaradha Ram
K. Dasaradha Ram
DRD. Shiva Reddy
D. Shiva Reddy
ASAmjan Shaik
Amjan Shaik
KRK. Ruth Ramya
K. Ruth Ramya
For most higher-level face evaluation applications, including liveliness, human activity identification, and personal contact, facial organization is a significant task. The real-world usefulness of such models is constrained since the utilization of present approaches can significantly deteriorate when handling images under particularly uncontrolled circumstances. This is true even though the best-in-class accuracy has significantly improved thanks to new access to large datasets and potent deep learning algorithms. In this study, a composite recurrent tracker that can simultaneously find single image facial arrangement and deformable facial following in nature, has been suggested. The multi-facet LSTMs are combined to illustrate real-world scenarios with varied length, and an internal denoiser that focuses on enhancing the information images to increase the resiliency of the general model, is offered. Face positioning is important in the majority of face examination frameworks. It emphasizes locating a few prominent features of human faces in images or recordings. The planning strategies and implementations described in this research, are based on information expansion and programming enhancement techniques, and they allow for working on a wide range of models with a place for specific continuous computations for face arrangement. A sophisticated set of evaluation metrics that enables new assessments to lessen the frequent issues seen in actual opportunity-following contexts, is proposed. The exploratory results show that the models created utilizing the approaches are more accurate, faster and more robust in defined testing environments, and more flexible in global positioning frameworks is the proposed challenge.

31 December 2025
Verified
Book Chapter
Open access
PDF file
JRJyothi Reddy
Jyothi Reddy
ASAmjan Shaik
Amjan Shaik
A Fusion Framework power-driven by intelligent technologies, designed for personalized and smart learning with multimedia capabilities aims to overcome the limitations of conventional one-size-fits-all educational models. It uses Audiotext Net for precise speech recognition, a next-generation intent recognition module, and Sub-Intent-Net to classify sub-intents accurately. These components enable rich, context-aware user interactions and adaptive and immersive learning experiences. The introduced system utilized many of the newest profound learning advancements, such as high-order ConvNet, Bi-GRU network, and BERT-based embedding to attain robust STT, intent classifier, and sub-intent classifier. We plan to extend the framework for multi-lingual communication, for emotion detection, and to deploy it on embedded devices to ensure access to global learning contexts.


9 March 2022 in Wireless Personal Communications
Verified
Journal Article
Open access
PDF file
PVPrasad Velpula
Prasad Velpula
RPRajendra Pamula
Rajendra Pamula
PJPraphula Kumar Jain
Praphula Kumar Jain
ASAmjan Shaik
Amjan Shaik

31 December 2025
Verified
Book Chapter
Open access
PDF file
BDBoddupelli Durgabhavani
Boddupelli Durgabhavani
ASAmjan Shaik
Amjan Shaik
In the meantime, the standard approaches predominantly focus on extracting spatial features from pre-trained convolutional neural network (CNN) architectures, including VGG16, InceptionV3, and DenseNet121, while rarely considering the sequential dependencies intrinsic to histopathological patterns. This constraint hinders their ability to represent intricate dependencies among image attributes, leading to inaccurate classification performance. This paper proposes a new hybrid deep learning framework (ColoSeqNet) for colorectal cancer detection to tackle these challenges. The proposed approach leveraged an enhanced ResNet 50 model integrated with long short-term memory (LSTM) layers for powerful spatial feature extraction and sequential learning ability. We use the improved ResNet50 for fine-tuning to fit a model on the colorectal histopathology images and LSTM layers for preserving sequential correlations and dependencies among extracted features. The ColoSeqNet architecture is validated on the open-to-access Colorectal Cancer Histology dataset. Results show that our model achieves an outstanding accuracy of 96.89% compared to traditional pre-trained models and recent state-of-the-art approaches.