search

abhisheks008 / DL-Simplified

Deep Learning Simplified is an Open-source repository, containing beginner to advance level deep learning projects for the contributors, who are willing to start their journey in Deep Learning. Devfolio URL, https://devfolio.co/projects/deep-learning-simplified-f013
https://quine.sh/repo/abhisheks008-DL-Simplified-499023976
MIT License
390 stars 339 forks source link

Payments Fraud Detection using DL #386

Closed Avdhesh-Varshney closed 9 months ago

Avdhesh-Varshney commented 11 months ago

Deep Learning Simplified Repository (Proposing new issue)

:red_circle: Project Title : Payments fraud detection model
:red_circle: Aim : Creating a ML project to detect any fraud in the online payments/transactions to secure the bank and users.
:red_circle: Dataset : https://www.kaggle.com/datasets/ealaxi/paysim1
:red_circle: Approach : Try to use 3-4 algorithms to implement the models and compare all the algorithms to find out the best fitted algorithm for the model by checking the accuracy scores. Also do not forget to do a exploratory data analysis before creating any model.

📍 Follow the Guidelines to Contribute in the Project :

:red_circle::yellow_circle: Points to Note :

:white_check_mark: To be Mentioned while taking the issue :

Happy Contributing 🚀

All the best. Enjoy your open source journey ahead. 😎

Avdhesh-Varshney commented 11 months ago

@abhisheks008 please assign this project to me to work on it and contribute it under Codepeak'23.

/assign

abhisheks008 commented 11 months ago

Hi @Avdhesh-Varshney it's a nice issue/project to work on. But this project repository mainly focuses on DL projects than ML projects. What are DL methods you wanna use for this project?

abhisheks008 commented 10 months ago

Hi @Avdhesh-Varshney are you still willing to work on this issue?

Nithish-456 commented 10 months ago

Please assign this project to me for contribution under SWOC 2024

abhisheks008 commented 10 months ago

Hi @Nithish-456 nice to have you here. Can you tell us a brief about the approach you are going to take for solving this issue?

Nithish-456 commented 10 months ago

Yes sir, My approach is to detect (classify) the fraud transactions in financial banks using 3 or 4 Machine Learning algorithms like random forest classifier, ensemble methods, support vector classifier and so on. So, by employing that algorithms on the provided dataset. I can able to recognize which algorithm is best ... and also we can use feature engineering, feature selection techniques further..

abhisheks008 commented 10 months ago

As this project repository majorly consists of deep learning projects, I suggest you look out for different DL methods/approaches for this project rather than going for basic ML methods. I am looking forward to hear from you.

nilam576 commented 10 months ago

I would like to take up this issue Full name : Nilam Bhojwani GitHub Profile Link : https://github.com/nilam576 Email ID : nilambhojwaningp@gmail.com Participant ID (if applicable): Approach for this Project : 1.Anomaly Detection with Autoencoders

  1. Generative Adversarial Networks (GANs) for Adversarial Training: 3.Recurrent Neural Networks (RNNs) for Sequence Patterns

What is your participant role? (Mention the Open Source program) - SWOC'24 contributor.

Nithish-456 commented 10 months ago

As this project repository majorly consists of deep learning projects, I suggest you look out for different DL methods/approaches for this project rather than going for basic ML methods. I am looking forward to hear from you.

Sir, in case of DL we can use MLP classifier or building a neural network. But my point of view is there is no sequential data to process, only we can do classification and also deep learning works good for larger datasets... So, if you want to do build a neural network I will go with DL approach sir. Or else I can choose another project issue... I am looking forward to hear suggestion from you. Thanks

abhisheks008 commented 10 months ago

Sorry of the inconvenience caused by me. Can you find out a dataset which matches with the issue's goal and can work good with neural networks?

Madhusri02 commented 10 months ago

Full name : Madhu Sri GitHub Profile Link : https://github.com/Madhusri02 Email ID : madhusri0774@gmail.com What is your participant role ? : Contributor at SWOC 2024 Approach for this Project : I will use LSTM , CNN , autoencoders to find the tricky patterns and irregularity in transactions.

dsid271 commented 10 months ago

Full name : Sidhartha Dondapati GitHub Profile Link : https://github.com/dsid271 Email ID : dsid271@gmail.com Participant ID (if applicable): Approach for this Project :

  1. Dataset Overview:

    • Utilized the PaySim synthetic dataset, simulating mobile money transactions from an African country.
    • Dataset size: 470 MB, 6,362,620 rows.
    • 11 attributes, target column: 'isFraud'.
    • Attributes: 5 numeric, 4 categorical, 2 string.

  2. Data Preprocessing:

    • Imbalanced data: 98.87% non-fraud, 0.12% fraud transactions.
    • Applied Synthetic Minority Over Sampling Technique (SMOTE) to balance the dataset, increasing the percent of the minority class.
    • Dropped 'step' attribute due to no correlation with transactions.
    • Removed 'nameOrig' and 'nameDest' (unique values) and 'isFlaggedFraud' (no impact).

  3. Feature Engineering:

    • Dropped unnecessary attributes for model training.

  4. Modeling Approach:

    • Utilized a deep learning approach with a Sequential neural network.
    • Applied standardization to numerical features and one-hot encoding to categorical features.
    • Implemented dropout regularization to prevent overfitting.
    • Used the Adam optimizer with a learning rate of 0.001.
    • Trained the model for 10 epochs with a batch size of 64.

  5. Evaluation:

    • Evaluated the model using accuracy as the performance metric.
    • The model's accuracy was assessed on the test set after training.

What is your participant role? (Mention the Open Source program): SWOC 2024

dsid271 commented 10 months ago

I used a feedforward neural network, an LSTM-based model, and an ensemble model that combines both for a more robust fraud detection system. I already completed worked on this project due to my personal interest

AgrawalTitiksha commented 10 months ago

Full name : Titiksha Agrawal

GitHub Profile Link : https://github.com/AgrawalTitiksha

Email ID : agrawaltn2311@gmail.com

Approach for this Project :

  1. Import necessary libraries (numpy, matplotlib, seaborn, pandas, tensorflow, keras, openCV)

  2. Performing necessary pre-processing (checking for missing values, duplicates)

  3. Visualizations for visual understanding of dataset.

  4. Creating all the required functions.

  5. Split data into test and train dataset.

  6. Train the model on Decision tree classifier, k nearest, and many test on many other algorithms. (ML algo, & can try working on converting non-image dataset to image and then apply CNN algorithm)

  7. Predict the test data and checking the accuracy.

  8. At the end, labeled all the accuracy of all models in a table.

What is your participant role? (Mention the Open Source program) : @contributor SWOC

abhisheks008 commented 10 months ago

Hi @AgrawalTitiksha focus on deep learning methods specially rather than machine learning models, as this project repository demands deep learning projects.

Issue assigned to you @AgrawalTitiksha

dsid271 commented 10 months ago

Hey @abhisheks008 was my approach not upto your expectations? Can you let me know where I went wrong?

abhisheks008 commented 10 months ago

Hey @abhisheks008 was my approach not upto your expectations? Can you let me know where I went wrong?

Hi @dsid271 your approach needs to be to the point. Eg. I want to use VGG, ResNet, MobileNet for this project. I have assigned to Titiksha, because in the last project she did a good job with the models and I expect she will upgrade her previous work. There are lots of issues open here, you can choose any of those to start work on.

dsid271 commented 10 months ago

Noted!