Spotting a phony attack by concealing the deception of the web was not the primary goal...the major purpose of this study was to determine challenging task of identifying fake online attacks where the attackers cleverly hide their true intentions within the website itself. APIs analyse request-response to sniff out phony URL disguise. The result of fake product journey tracks using block chain exposing fakes at a glance. Even fake reviews have been detected using sentimental analysis where AI sniffs out words to expose hidden review algorithms shift data, guarding against fraudulent transactions.
The overall view of this paper is, spotting online fraud involves highly skilled and smart tools, URLs flagged for plagiarism, reviews examined for suspicious tone, products tracked on block chains, and payments detected using effective algorithms as shown in fig 1.
3.1 Fake Url’s
This study looked at and detailed a system that uses N-gram models, request-response APIs, and Support Vector Machines (SVMs) to detect malicious URLs.
3.1.1 Motivation:
Malicious URLs pose a significant threat to cybersecurity, enabling malware attacks and data breaches. Early detection of these URLs is crucial for user protection.
3.1.2 Proposed Approach:
Utilize N-gram models to extract features from the URL structure, capturing patterns and sequences of characters. Employ SVMs, a machine learning algorithm, to classify URLs as malicious or benign based on the extracted features.
3.1.3 Justification For Using SVMs:
SVMs offer high accuracy for large datasets, suitable for handling extensive URL lists. They excel at text classification tasks like identifying malicious patterns. Malicious URL Detection with N-gram and svm.
3.1.4 SVM:
Focuses on analysing the URL structure itself (characters, sequences) using N-grams. Employ’s machine learning (SVM) to classify URLs as malicious or benign based on extracted features. Doesn't directly involve API requests or responses.
3.1.5 Spotting Deceptive Web Attacks:
Often targets APIs as entry points for injecting malicious code or data. Involves analysing API request-response patterns, data payloads, and potential vulnerabilities. Doesn't directly utilize N-gram or SVM techniques. While both areas aim to enhance cybersecurity, they address different aspects of web threats.
3.1.6 Implementation Steps:
Pre-Processing:
Tokenize URLs into individual words or characters. Extract N-gram features (e.g., consecutive pairs or triplets of characters).
Training:
Train the SVM model on a labelled dataset of known malicious and benign URLs. The model learns to distinguish between the two classes based on the extracted features.
Prediction:
Present new URLs to the trained model. The model predicts whether each URL is likely to be malicious or not.
Results:
The authors achieved the best results using the N-gram and SVM approach compared to other methods.
3.2 Fake Product
Here's the next module for this paper: Ganache has introduced a smart contract for Ethereum, enabling secure Dapp development, launch, and testing, while a MetaMask cryptocurrency wallet is used for transactions. Dapp users can't log in without first authenticating. Manufacturers join the DApp with ID by enrolling products with price, QR code, and serial number. Customers use a QR code or serial number to verify items, sellers submit a seller code, and data is stored on a blockchain network .
3.2.1 Block chain
Supply chain management can benefit greatly from blockchain technology's ability to store data on nodes, which allows for the tracking, sharing, and security of accounts, payments, orders, and product prices. Some features of this technology include:
1.Security and privacy:
Blockchain ensures users have a public and private key combination for transaction validation, guaranteeing immutable and permanent transactions. Blockchain uses public key encryption to secure data.
2.Decentralization:
Blockchain is a distributed ledger technology that functions independently of outside parties or a centralized authority.
3.Transparency:
Smart contracts control the public access to blockchain data. The suggested solution involves Manufacturer, Seller, and Consumer stakeholders and makes use of Ganache Ethereum's and the MetaMask wallet.
3.2.2 Ethereum Dapp Architecture
The user interface of the system is based on NodeJS, and Web3.js establishes a connection with MetaMask, a provider, to generate and certify transactions using the user's private key. This ensures safe network communication without storing the private keys of users.
The blockchain's distributed ledgers are transforming the ownership tracking system. The business and e-commerce industries are changing quickly, which is having an impact on supply chain trends. The Dapp created here may be trusted to be used in e-commerce and guarantees more openness in supply chain management. We have exclusively examined medical products in this study.
3.3 Fake Review
The present study investigates the efficacy of machine learning to detect fraudulent e-commerce evaluations by concentrating on linguistic and affective factors. The objective is to improve transparency by identifying characteristics of fraudulent reviews, using an Amazon dataset encompassing 21,000 reviews that have been equally divided into real and fraudulent reviews.
3.3.1.Data Preparation
Further steps in the implementation and data preparation process might be necessary, such as controlling punctuation and special characters and applying cutting-edge methods to improve the sentiment analysis model's performance.
3.3.2. Feature Extraction
Feature extraction is a method that produces better results than directly applying machine learning to raw data because it transforms unprocessed input into numerical features.
3.3.3. Feature Selection Code Using The RFE
Recursive Feature Elimination (RFE) is a machine learning technique that ranks features based on relevance, removing the least important ones until the required number of features is achieved.
3.3.4.Sentiment Classification Algorithms.
Several well-known classifiers will be used in this study:
1) Naïve Bayes (NB)
A simple probabilistic classifier based on the Bayes theorem is the NB classifier. A set of probabilities is computed by the NB using combinations of values from a given dataset.
Additionally, the NB classifier makes decisions quickly.
2) SVM, Or Support Vector Machine
Despite the development of new techniques, SVM is a popular supervised learning model in machine learning, evaluating data to uncover patterns for regression analysis and classification.
3) Logistic Regression
Although the name suggests otherwise, this technique is applied to classification problems. It's a straightforward yet effective sentiment analysis method.
4) Random Forest
To increase accuracy and performance, this ensemble learning technique mixes many decision trees.
5. Detection Process And Result
Following training, the model's output on the testing dataset is predicted, and a confusion matrix is created to categorize the reviews as positive or negative.
This study emphasizes the need for greater awareness and comprehension of online reviews by examining the use of textual traits, mood, and linguistics to identify and forecast fraudulent reviews.
4.4 Fake Payment
The most widely used form of payment in the world today is online. But as the number of online payments rises, so does the incidence of payment fraud. This study aims to distinguish between fraudulent and non-fraudulent payments.
4.4.1 Data Preprocessing:
Data loading into a pandas Data Frame, handling missing values, fraud detection feature engineering, categorical variable encoding, and data splitting into groups for testing and training is every phase in the process
4.4.2 Exploratory Data Analysis (EDA):
To begin comprehend variable distribution, this task involves creating descriptive statistics. Data distribution and fraud patterns then become apparent through the use of visualizations.
4.4.3 Feature Selection:
Correlation analysis is used in the process to find significant predictors, and feature importance techniques were performed to select relevant behaviors.
4.4.4 Model Building:
The Random Forest algorithm is used to detect fraud. Performance metrics are determined, hyperparameters change, and a dataset is employed to train the model.
4.4.5 Model Interpretation:
In an effort to discover areas for improvement, the research comprises studying the model's predictions, establishing the factors influencing its evaluation, and looking at misclassifications.
4.4.6 Deployment:
The trained model is to be involved into the present setup, watched and retrained on frequently, and a feedback loop should be set up to collect information pertaining to fraud cases that have been identified.
4.4.7 Continuous Improvement:
The necessity of remaining up with the most recent fraud detection tools and incorporating feedback from ongoing monitoring and user interactions is repeatedly highlighted during the text.