2023

Permanent URI for this collection

https://escholar.umt.edu.pk/handle/123456789/6866

Browse

Now showing 1 - 20 of 26

False data injection attack detection using machine learning
(UMT, Lahore, 2023) WAFA NOOR
With the rapid integration of smart grid technologies and increased reliance on communication networks, the security of power systems has become a critical concern. False Data Injection (FDI) attacks pose a significant threat to the stability and reliability of smart grid operations. This research aims to evaluate a novel approach for FDI attack detection in smart grids using machine learning techniques. The research explores various machine learning multi-label algorithms, such as Label Powerset, Classifier Chain, Binary Relevance, and ML-KNN, to analyze power system data and identify anomalies indicative of FDI attacks. The 14-bus power system and simulated attack scenarios are utilized to evaluate the proposed approach’s effectiveness and robustness. Results were evaluated based on row accuracy, precision, recall, and F1-score. ML-KNN outperforms all the other methods by giving 99.1% row accuracy. The findings of this research contribute to enhancing the security of smart grids against cyber threats and lay the foundation for further advancements in FDI attack detection using machine learning methodologies.
ANTIMICROBIAL PEPTIDES PREDICTION USING MACHINE-LEARNING ALGORITHMS
(UMT, Lahore, 2023) ROBBIA GULNAR
Evolution of drug resistant microbes develop interest among scientists in developing, discovering and re-engineer antibiotics. Antimicrobial resistance is major global health problem to combat with the issue there is need of antimicrobial treatment such as development of antimicrobial peptides. AMPs have received increasing attention as a treatment for several infectious diseases, including bacterial infections, viral infections, and fungal infections. They have several advantages over conventional antibiotics, including broad-spectrum activity, the potential to reduce resistance, and the ability to kill microorganisms quickly. However, there are several challenges associated with the development of AMP-based therapeutics, including stability, toxicity, and manufacturing cost. Despite these challenges, the potential benefits of AMPs make them an attractive research area for the development of new antimicrobial agents. Fortunately, AMP sequences with potential bactericidal properties have evolved naturally because of competition within complex communities. Unfortunately, the discovery, description, and production of AMPs is difficult and time-consuming. In this study, we used dataset of 2638 AMPs and 3700 non-AMPs to build model for AMP Prediction. We explore nine various models of “machine learning are Random Forest, Decision Tree, Naive Bayes (NB), and Support Vector Machine”, Quadratic Discriminant Analysis, Linear Discriminant Analysis, Extreme Gradient Boosting, k-Nearest Neighbor, and Light Gradient Boosting Machine (LGBM). In comparison to previous models, the LGBM classifier outperformed than all with an accuracy rate of 93.85%, and properly predicting both positive and negative datasets of AMPs. Our discussion of machine learning predictions of antimicrobial peptides concludes with developments, limitation, and potential future approaches.
DEEP LEARNING METHODS FOR PV POWER FORECASTING
(UMT, Lahore, 2023) RABBIA IJAZ
This study focuses on the precise forecasting of photovoltaic (PV) power, considering the intermittent and weather-dependent characteristics of solar energy. However, obtaining weather data and atmospheric measurements can be prohibitively expensive, particularly in developing nations. To address this challenge, we propose an innovative approach that utilizes deep learning-based prediction models solely relying on PV power data for accurate forecasting. Specifically, we investigate the application of two potent deep learning architectures, namely Long Short-Term Memory (LSTM) and Convolutional Neural Networks (CNN), in PV power prediction. LSTM effectively models and captures long-term dependencies in time series data, while CNN excels in extracting spatial features. Furthermore, we propose an ensemble model that combines LSTM and CNN predictions through a weighted average approach, leading to improved accuracy and predictive power compared to individual models. This ensemble methodology offers a practical and cost-effective solution for PV power forecasting in developing countries, where weather data may be limited, thus facilitating the seamless integration of renewable energy resources into the power grid.
A NOVEL CUBIC CHAOTIC MAP FOR SBOX DESIGN AND CIPHER
(UMT, Lahore, 2023) MUHAMMAD AHMAD
The rapid advancements in technology have resulted in a significant rise in the quantity of data being shared and transmitted across various sources on the network. With data being a critical aspect of modern life, its importance cannot be overstated. Unfortunately, this also makes it easier for hackers to gain access to and steal sensitive information. To maintain data security and protect against such threats, secure connections are essential. Cryptography is employed to enhance the security of private data and communication networks, utilizing various substitution and permutation strategies. Many researchers have developed algorithms using dynamic substitution boxes to bolster security against attackers. However, since some of these S-boxes are weak and ineffective, there is a constant need for strong S-Boxes. In this thesis, a robust key-based substitution box (S-Box) based on the 1-D chaotic method is proposed and its strength using standard benchmarks has been evaluated. The results are promising that indicate that the proposed S-Box could be utilized in modern ciphers with confidence. A new cipher has been designed to use the proposed S-Box to provide security of data. The proposed cipher has eight rounds and each round is performing the same operations. The obtained results of our proposed S-Box are comparatively worthy. The cipher's proposed plaintext size is 256 bits, while the key length is 256 bits.
DESIGN OF A RADICAL CHAOTIC MAP S-BOX AND A NOVEL ENCRYPTION METHOD
(UMT, Lahore, 2023) SHAHEER ALI
Information is a vital resource that drives the functioning of numerous systems and processes in the modern digital era. Routinely transferred across various platforms via a variety of communication routes, information serves as the backbone of our interconnected world. However, such widespread information interchange also exposes the data to the constant risk of loss and malicious attacks. This study offers a unique encryption that boosts data security by using a chaotic substitution box. The substitution box is one of the key elements of the cipher, crucial for creating chaos in the data. An encryption algorithm is required to reduce these hazards. A substitution box is used in a cipher to create confusion and protect the data. In this research, chaotic formula issued to build a substitution box. Furthermore, a novel encryption method is presented, which encrypts 256 bits of data using the proposed substitution box. The cipher uses 128 bits of key to accomplish several operations across its 16 rounds of encryption. A unique key is created for encryption for each round. Due to its chaotic behaviour and the proposed substitution box's serious complexity, the cipher offers a high level of encryption standard. The produced substitution box is analysed using several metrics, and the proposed cipher undergoes extensive testing to validate its efficacy. The outcomes show that the suggested encryption provides better data security when compared to conventional ciphers. Likewise, the suggested cipher is adaptable and can be changed to meet specific needs.
A CIPHER PROPOSAL BASED ON A NOVEL TRIGONOMETRIC CHAOTIC MAP
(UMT, Lahore, 2023) UMER NAWAZ
In the ever-evolving digital era, the sharing of information has become commonplace across numerous platforms and channels. However, this widespread exchange also exposes data to potential vulnerabilities and threats. To counteract such risks, the implementation of a cipher algorithm becomes imperative. Modern ciphers utilize substitution boxes to introduce chaos and fortify data security. In this study, a unique substitution box is constructed utilizing trigonometric functions. To assess the effectiveness of the substitution box, its properties are thoroughly evaluated. Additionally, a novel block cipher is introduced, utilizing the substitution box to encrypt 128 bits of data. The cipher consists of 16 encryption rounds and necessitates a 128-bit key for executing multiple operations. Each round involves the generation of a distinct key for encryption purposes. Comprising four primary functions and a pre-function, the cipher incorporates the designed substitution box to offer substantial complexity, thus ensuring a high level of encryption standard through its chaotic behavior. To validate the efficiency of the proposed cipher, the constructed substitution box undergoes meticulous evaluation using various metrics, while the cipher itself is extensively tested. The obtained results unequivocally demonstrate that the proposed cipher outperforms traditional ciphers, significantly enhancing data security. Furthermore, the proposed cipher exhibits scalability and can be tailored to specific requirements, thereby offering versatility in its application.
PREDICTION OF PROLIFERATIVE DIABETIC RETINOPATHY USING BIDIRECTIONAL LONG-SHORT-TERM-MEMORY BASED CONVOLUTIONAL NEURAL NETWORK
(UMT, Lahore, 2023) MUHAMMAD BILAL KHAN
Proliferative diabetic retinopathy (PDR) is a severe complication of diabetes that can lead to blindness. Early detection and treatment of PDR are crucial to prevent blindness. In this study, a deep learning-based model is proposed to predict PDR using bidirectional long short-term memory (LSTM) based convolutional neural networks (CNNs). Diabetes has a serious consequence called proliferative diabetic retinopathy (PDR), which if uncontrolled, can result in blindness. Effective management of PDR requires early discovery and diagnosis. Ophthalmoscopy and fluorescein angiography, two common traditional diagnostic techniques, take time, need training, and sometimes cause discomfort for the patient. A quicker, more automatic, and less intrusive method of detecting PDR, however, can be provided by deep learning techniques. For the purpose of predicting PDR using retinal fundus images, a method of deep convolutional neural networks Based on Bi-LSTM was built and evaluated in the current study. On a sizable dataset of pictures from patients with and without PDR, the suggested model was trained and put to the test. 450,000 patches of fundus images from the two datasets were used for training and testing. The experimental findings demonstrated that the suggested Bi-LSTM-CNN model outperformed other cutting-edge techniques. The outcome accuracy of the model was 98.4% respectively. These findings show the suggested model's potential as a helpful tool for the early detection and diagnosis of PDR. Retinal fundus images can be used to improve the diagnosis of PDR using the proposed deep learning approach based on Bi-LSTM-CNN. The automated method might make healthcare workers less stressed, enhance patient outcomes, and ultimately lower the incidence of PDR-related blindness. The current study demonstrates the potential of deep learning-based models to improve the early detection and treatment of PDR. Recognition of abnormal retinal vessels which leads to tortuosity enables us to diagnose multiple retinal diseases. Tortuosity can be interpreted as the high curve value along with the multiple segments of vessels. The proposed algorithm effectively specified the tortuosity of vessels based on curvature and width/thickness of vessels.
ENHANCING TWITTER CONTENT MODERATION: A SOFT VOTING CLASSIFIER FOR OFFENSIVE AND HATEFUL TWEETS
(UMT, Lahore, 2023) HUSSAIN SHAHID
The widespread use of social media, such as Twitter, has transformed global communication but has also led to a challenge: the presence of harmful and discriminatory content. This study introduces a Soft Voting Classifier model to enhance Twitter content moderation by accurately identifying offensive and hateful tweets. This model employs ensemble learning, combining deep learning, traditional machine learning, and rule-based systems to thoroughly assess tweets and improving the classification process. Effective model performance relies on feature selection. Identifying offensive and hateful content demands a deep understanding of language and context. The model uses a hybrid approach, integrating linguistic features like n-grams and sentiment scores with advanced features like word embeddings and contextual embeddings, considering both explicit and implicit indicators. To optimize the model's performance, careful fine-tuning involves hyperparameter selection, cross-validation, and training on a large dataset. Comprehensive evaluation metrics like precision, recall, F1-score, and accuracy gauge its effectiveness in identifying abusive tweets. This model exhibits robust performance with an accuracy of 0.94 on individual dataset and on the combined dataset. In summary, this study presents a Soft Voting Classifier model tailored for Twitter content moderation. By leveraging multiple base classifiers and advanced feature selection, it offers the potential to create safer online environments, promoting responsible digital interactions. As the digital landscape evolves, a heightened emphasis on content moderation technologies becomes essential, reducing harmful content and fostering a more inclusive digital space for Twitter users.
AN ENCRYPTION SCHEME BASED ON LOGARITHMIC SQUARE POLYNOMIAL
(UMT, Lahore, 2023) HINA ZAFAR
In today's digital era, information is a valuable resource that is routinely transferred among many resources through various communication methods. Such information exchange, however, exposes the data to possible assaults and loss. A cipher algorithm is required to reduce such hazards. A substitution box is used in ciphers to produce chaos and safeguard information. This study describes a new cipher that uses a chaotic substitution box to improve data security. A substitution box is produced in this study utilizing trigonometric functions. The parameters of the substitution box are used to assess its robustness. A novel block cipher is also developed, which leverages the suggested substitution box to encrypt 128 bits of data. The cipher has 8 encryption rounds and requires 128 bits of key to accomplish multiple operations. For encryption, a new key is created for each cycle. Because of its chaotic behavior, the suggested substitution box is meant to give great complexity, and the cipher delivers a high degree of encryption standard. To validate the suggested cipher's efficacy, the created substitution box is assessed using several metrics, and the cipher is fully tested. The results show that the suggested cipher provides better data security than standard ciphers. The enhanced S-box and the suggested encryption technique work together to deliver increased security while preserving computing efficiency, demonstrating a substantial advancement in data protection paradigms.
DEVELOPMENT OF A NOVEL CIPHER BASED ON AN INNOVATIVE NON-LINEAR COMPONENT
(UMT, Lahore, 2023) MUHAMMAD HAZEEL AHMED
Security of data is the major part of society and with day to day life it is becoming the need of today’s world. To facilitate the society, different modes of securing the information are being used. Cryptography is used to make personal information more secure by using different ways to change and rearrange data. Different ciphers have been introduced with various methods and algorithms to secure the data. Modern ciphers are using Substitution Box to provide the security of such data. The working of substitution box is to make data meaningless for an attacker. In this article we have proposed a new chaotic map in order to design an innovative cryptographic cipher with the help of a new and strong S-box by adopting modular functions on the other hand comparing it with NL, FP, SAC, BIC, LP and DU of other S-boxes. The approach towards innovative chaotic map is furthermore efficient in the production of the stronger S-Box rather than others. Our proposed substitution box values are familiar and close to the other S-boxes.
AN ENCRYPTION ALGORITHM BASED ON AN INNOVATIVE CHAOTIC MAP
(UMT, Lahore, 2023) MUHAMMAD ANS
Many companies possess data or information that only a select group of authorized individuals can evaluate. This information may be owned by a country's telecom industry, software firms, financial institutions, or the military. As the world moves towards digital growth, this dataset is being converted into a digital format, but it comes at a cost. While online information is reasonably secure, cyberattack development is also advancing. Therefore, cryptography and its numerous techniques are employed to ensure data security and integrity. Substitution boxes and permutation algorithms are frequently used in cryptography to safeguard sensitive data. S-boxes, due to their dynamic nature, are a suitable choice for security purposes in the present day. In this thesis, a simple yet effective cipher with three basic components is constructed. These components include key creation, the encryption procedure, and decryption. For the reader's convenience, their schemas and examples are also provided. After creating the S-Box is assessed using non-linearity, BIC, SAC, LP, and DP S-Box attributes. The outcomes demonstrate that the suggested cipher is robust enough to be applied to real-time applications
A RADICAL CHAOTIC MAP-BASED S-BOX AND ITS APPLICATION IN A NEW CIPHER
(UMT, Lahore, 2023) KAINAAT MALIK
The broad usage of communication technology nowadays suggests the necessity for certain security measures to be put in place. Different ciphers are being created to do this by utilizing numerous strategies. These ciphers are currently being created using chaotic maps. Using chaos in cryptographic algorithms has the advantage of making data more unstable and inconsistent. A substitution box is a unique and nonlinear essential component of block ciphers. To fend off modern attacks, extremely diffusive S-Boxes are required as nonlinear confusion sublayers in cryptosystems. By adopting better substitution box design approaches, a ciphertext of higher quality can be made. In this research, a novel chaotic map is suggested and used to create a new S-Box. A new cipher is created employing the suggested S-Box and a new function. Along with the decryption procedure, the key generation approach for the cipher is also described. Nonlinearity (NL), Linear Approximation Probability (LP), Differential Approximation Probability (DP), Strict Avalanche Criteria (SAC), and Bit Independence Criteria (BIC) are among the parameters taken into consideration when analyzing and evaluating the proposed S-Box's effectiveness against various cyberattacks. The results of the performance test show that the suggested S-Box has strong cryptographic characteristics, is resistant to several attacks and is appropriate for use in modern ciphers.
DETECTING MULTI-LABEL ATTACKS IN AC MICROGRIDS BY USING MACHINE LEARNING
(UMT, Lahore, 2023) Saman Liaqat
In today's competitive market, power utilities are increasingly turning to innovative smart grid technologies to improve the efficiency of their grids. However, this comes with a heightened risk of cyber-attacks, making strong cybersecurity measures essential for the protection of power grids. To detect and counteract false data injection (FDI) assaults, this study suggests a multi-label attack detection method for AC microgrids. In the secondary control of AC microgrids, the method uses frequency and voltage control variables as an optimization problem. The task is formulated as a multi-label classification problem to overcome the difficulty of finding the co-occurrence dependencies and discrepancies in power flow measurements brought on by FDI attacks. This allows for the use of a single model to detect various types of attacks and changes in load, improving the efficiency of the attack detection process. The proposed technique is compared against five different machine learning approaches, and its performance is evaluated using the IEEE 34-bus distribution test system. The results demonstrate effectiveness of the multi-label attack detection technique in identifying and mitigating cyber-attacks in AC microgrids. Overall, this paper provides valuable insights into the development of effective cybersecurity measures for power utilities, assisting to guarantee the secure and reliable operation of electricity grids in today’s competitive market.
SOCIAL MEDIA FAKE ACCOUNTS AND NEWS DETECTION USING MACHINE LEARNING
(UMT, Lahore, 2023) SAIRA SATTAR
Social media platforms have become integral parts of our daily lives, providing avenues for communication, information sharing, and social interaction. However, with the widespread use of these platforms, the emergence of fake accounts and the dissemination of fake news has become a critical concern. Fake accounts refer to profiles created with deceptive intentions, often with the aim of spreading misinformation, manipulating public opinion, or engaging in illicit activities. Meanwhile, fake news involves the deliberate creation and circulation of misleading or fabricated information presented as legitimate news. But in some cases, few fake clients deliver unique user some off base recommendations. They attempt to bug genuine client or attempt to recover Account information of genuine clients. For that reason, we are propelled to do this proposal to solve the issue by recognizing fake user account applying Machine Learning Algorithms
A BLOCKCHAIN BASED FRAMEWORK FOR POTATO SEED TRACEABILITY IN AGRICULTURE SUPPLY CHAIN MANAGEMENT
(UMT, Lahore, 2023) FATIMA NAWAZ
DISTANCE LEARNING TECHNICAL & NON-TECHNICAL ISSUES AND THEIR SOLUTIONS
(UMT, Lahore, 2023) Yasmeen Farid
Distance learning is any form of remote education where the student is not physically present in the class. There were many reasons to start distance learning all over the world. The main purpose of which was to give people in remote areas access to higher education and educating people who for some reason couldn't take physical classes. Today, thanks to Distance learning, you can easily read any subject and learn any skill. In those special days of COVID-19, we had no choice but to move towards online education. There is a lot of damage to our education in the Special COVID-19 pandemic Situation anyway, but if we did not adapt to online education, there would be more damage. There are many advantages of online education, but there are also we discuss some technical issues. Distance learning is any form of remote education where the student is not physically present in the class. There were many reasons to start distance learning all over the world. The main purpose of which was to give people in remote areas access to higher education and educating people who for some reason couldn't take physical classes.
SMART GRID DATA SECURITY AND USER PRIVACY CONCERNS THROUGH BLOCKCHAIN TECHNOLOGY
(UMT, Lahore, 2023) AZIB MAHMOOD
The world of power grids technologies is converging towards a new concept that is smart grid (SG) technology. The basic objective of smart grid technology is to design an autonomous network in which all types of energy resources, consumers, and any other entity related to electricity or energy are linked up together to enhance energy conservation. In contrast to traditional grid technologies, the smart grid has numerous benefits like the bidirectional flow of information, safety and reliability, efficiency, resilience, and sustainability of the energy network. By utilizing modern technologies, SG is turning into the next-generation grid. With all these blessings, SG is still in its nascent stage and gathering the goodness of related technologies to be more reliable and dependable. The integration of smart grid technology into the existing power grid infrastructure has the potential to improve energy efficiency, reliability, and cost-effectiveness. However, the collection and processing of large amounts of sensitive user data by smart grid systems pose significant privacy and security risks. Block chain technology has emerged as a potential solution to these challenges, offering a decentralized, tamper-resistant, and transparent mechanism for data management and protection. In this paper discuss integration of block chain within the smart grid to protect the user of the grid. Examines the block chain's system architecture and proposes a security diagnostics technique for the smart grid based on current issues with that system. In this paper also discuss and suggest direction of applying the block chain in the smart grid for the privacy protection of the consumers and to handle the security issues. The paper also highlights the benefits and limitations of using block chain technology for smart grid privacy protection and identifies future research directions in this area.
EdgeEnergy: A blockchain-based energy trading framework for smart homes & grids
(UMT, Lahore, 2023) Syed Arbaz Haider Gilani
The prevalence of crowd-scale distributed energy generation encourages the concept of peer-to-peer (P2P) electricity sharing among electricity prosumers (producer and consumer). This thesis explores ”Edge Energy” – a distributed electricity trading system over blockchain to facilitate P2P electricity trading. The proposed system orchestrates the P2P electricity distribution network as a hierarchical network of electricity prosumers, where each prosumer has capabilities for communication and energy routing. The first layer (i.e. core layer), presents the coalitions of micro-grids. The second layer is the electricity distribution network layer, i.e. a software-driven energy router’s network. Lastly, the edge layer presents the end-prosumers. Edge Energy is designed as a blockchain-assisted multi-agent system to support the prosumers network and an agent coalition mechanism to form coalitions and negotiate electricity trading. Considering the P2P nature of the proposal and the intrinsic trust deficit between the stakeholders, the market-related functionalities (i.e. matching, trading, and transaction settlement are being performed over the blockchain using smart contracts). Simulations are conducted based on the java agent development environment to validate the proposed electricity trading process.
Design of a Chaotic Nonlinear Component for a Novel Encryption Method
(UMT, Lahore, 2023) AYESHA ASHRAF
Computerized systems often hold large quantities of data. While the information is being processed, if adequate safety measures are not implemented, it might be stolen or used inappropriately. Failures in data management, including sending private information outside of secure settings, might result in data being misused. Data security is crucial to prevent this by shielding confidential data from unauthorized access by attackers. Many years ago, ciphers formed the foundation of cryptographic systems used to secure data. Key creation, encryption, and decryption are the three stages that make up a cipher. Researchers advise using S-Boxes, chaotic maps, and permutation functions in ciphers to enhance data security. There are many distinct ciphers, however as time passed, hackers discovered many techniques to decipher the ciphered material, necessitating the invention of new ciphers in the realm of cryptography. Using the use of trigonometric and algebraic functions, I have presented a cipher in this work that utilizes a novel encryption and decryption process using a chaotic map for the substitution box. The suggested cipher is nonlinear by nature, but I will attempt to make this cipher more secure and powerful to achieve better accuracy and outcomes.
Prediction of Phage Virion Proteins: A Deep Learning Approach
(UMT, Lahore, 2023) ALLAH DITTA
Accurate prediction of phage virion proteins (PVPs) can greatly aid in the development of antibacterial drugs. In this study, we compared the performance of three approaches for predicting PVPs: machine learning algorithms, a Long Short-Term Memory (LSTM) deep learning model, and the KerasTuner. The machine learning algorithms, which required the extraction of features from the primary sequence, were unable to outperform the SCORPIAN classifier and achieved an MCC of 45% on independent testing. The LSTM model achieved an MCC of 51%, but also failed to outperform SCORPIAN. Interestingly, the KerasTuner method did better than the other methods, with a high score of 0.9921 on independent testing. These findings suggest that the KerasTuner approach is a good option for predicting PVPs and helping with the development of antibacterial drugs. If you want to check the results, you can find the code on GitHub.

Browse

Recent Submissions