ClarissaDeWi
commented
2 years ago @com90 Thanks so much for submitting this proposal. 👏 It looks like you still need to complete part 2 of the application. Please complete this simple form (https://forms.gle/a124xJmzeWesPeWj7) to streamline the contract process. This form is the final step in the submission process and a lack of this information will delay the contract.
Project
Fraud detection in helium transactions using machine learning approach
Elevator Pitch
Like every crypto currency HNT will be suffering from fraud behaviors in the near future. There are already reports of fraud accounts and devices which undermines the trust to the network and causes unfair income for other network users. This project will use machine learning algorithms to the Helium network transactions to detect fraud behavior and monitor them.
Total fiat/HNT Ask
65.000 USD
Name & Address
Prof. Dr. Mitat Uysal - https://www.dogus.edu.tr/en/academic-staff/faculty-of-engineering/mitat-uysal Mustafa Ozan Uysal - https://www.linkedin.com/in/ozanuysal/ Ahmet Faruk Acar - https://www.linkedin.com/in/ahmet-faruk-acar-87b122a1/
Address : Barış Sokak Alya Life Residence No:1 Kat:10 Daire:119 Fikirtepe / Kadıköy İstanbul
Team or Project Web Site
Appcent - http://appcent.mobi
Appcent is an award-winning technology company specialized in fin-tech and e-commerce that builds mobile applications across the Turkish and Middle East market for many well recognized brands such as Mavi, xCite, Costa Coffee, Alghanim Industries, Hepsiburada, NetWork, Metro, Unilever. Our R&D department has successfully developed and managed innovative machine learning projects funded by TUBITAK (The Scientific and Technological Research Council of Turkey), as well as self-funded IoT projects. Furthermore, with Appcent Academy, we organize training and young talent programs to transfer our sectoral experience to the ecosystem. Most recently in 2021, we have founded Appcent Design to provide end-to-end technology solutions to our partners.
A team of 3 developers, one project manager and academic consultant will be involved in the project.
Project Details
The project will consist of three modules;
Monitoring and data collection module Fraud detection using machine learning Analytics and alert dashboard
In the first phase we will analyze transactions on the blockchain and train our model to detect fraud behavior. Fraud behavior can be; money laundering, unfair HNT gain, a transaction which the wallet owner is unaware of or any other fraudulent activity. After that we will try different models and measure the performance of algorithms. A dashboard will help users to identify suspicious transactions and raise alerts if necessary.
Technical Objectives
Detection of fraudulent actions in blockchain transactions using ML consists of 4 steps;
Input data Extract features Train algorithm Create model
To detect fraud, a machine learning model first needs to collect data. The model analyzes all the data gathered, segments, and extracts the required features from it. Next, the machine learning model receives training sets that teach it to predict the probability of fraud. Finally, it creates fraud detection machine learning models.
PROCESSING THE DATA
Processing the data is the most important part in ML.
It consist of the following steps:
Evaluation and analysis Obtaining data files File format types Preparation data analysis Arranging and organizing data Data Analytical tests
FRAUD DETECTION
Recognizing fraud Data analytical software Anomalies versus fraud within data Fraudulent data Inclusions and deletions
FEATURE EXTRACTION AND SELECTION
Extraction : Getting useful features from existing data Selection : Choosing a subset of the original pool of features
FEATURE SELECTION IN THIS PROJECT
In this project, a novel cost-sensitive metaheuristic algorithm called as Migrating Birds Optimization(MBO) is used for cost effective feature selection.MBO is developed by E.Duman,M.Uysal and A.F.Alkaya.The steps of MBO is given as below:
n = the number of initial solutions (birds) k = the number of neighbor solutions to be considered x = the number of neighbor solutions to be shared with the next solution m = number of tours K = iteration limit
Pseudocode of MBO:
THE MAIN ALGORITHMS USED
After selecting the feature, then we are ready to build a ML classification model since fraud detection is a classification problem.
If we look at the literature the following algorithms are used for classification;
Neural Networks(NN) k-Nearest Neighbor(kNN) Support Vector Machine(SVM) Decision Tree Random Forest AdaBoost
We will try all of them to obtain the best results.
Roadmap