GAN-Based Training for Binary Classifier is a data challenge organized in association with [IEEE BigData 2022 conference](https://bigdataieee.org/BigData2022/BigDataCupChallenges.html). The challenge is to use pictorial-based or raw data-based GAN to generate malign samples suitable for classifier training. The competition could accept 10 papers based on the competition results to be published in the conference proceedings. Those papers should be presented at the associated workshop at the IEEE Big Data 2022 conference organized by each Cup Challenge and will be indexed in the same way as regular conference papers.

## Background
Malicious software, also known as malware, threatens our cyber services by infecting computational devices like computers and mobile phones. Nowadays, cyber security companies are oriented towards Machine Learning (ML) technologies over static detection techniques as a promising solution to overcome detection limitations and enhance their services. According to a Securelist's article published on 1st of Mar. 2021; the security tools of a single supplier (Kaspersky) detected in a single year 5,683,694 malicious installation packages, 156,710 new mobile banking Trojans, and 20,708 new mobile ransomware Trojans. Today ML-based solutions' reliability is mostly evaluated by collecting real malware inferences within an actual platform environment and not an emulator.
Behavioral polymorphism is a product of malware developers’ attempts to avoid the existing detection mechanisms by changing the malware’s operational trace while keeping its main tasks intact. Hackers accomplish this by manipulating the malware code using multiple forms of obfuscation like payloads encryption, character code shifting of the variable name and register reassignments. For example, Agobot or Gaobot is a computer worm that was first identified in 2002.  Over 580 versions of this malware were subsequently identified, some of them behaviorally polymorphic. Today, each malware can mutate to generate hundreds of polymorphic versions, adding up to thousands for each malware family. Generative Adversarial Networks (GAN Systems) have been successfully used in solving computer vision, object detection, and classification problems by learning, mimicking, and generating any distribution of data. For example, they produced high-fidelity images visually indistinguishable from real ones or synthetic emails textually close to real messages. In the cyber security research field, one would like to use GANs to generate synthetic malware behavior, avoiding the challenge of manually collecting the training malware traces.

## The Challenge:
The challenge is to develop a generative adversarial network to generate a proper training set for the an ML–based malware classifier.  For our competition, you should generate malware behavior samples in the same format as the manually collected ones.
You will notice that the manually collected training set, which contains a histogram representation of malware and ordinary software behavior, is unbalanced (3000 benign and 1465 malign samples) hence, the challenge is to:
- Train an ML model using the unbalanced real dataset (train.csv) and submit the classification outcome on the validation dataset (Validation.csv) as to get the current RMSE feedback as a baseline. Only RMSEs below 0.2 will be considered.
- Develop a GAN system using the available real dataset as guidance to generate synthetic samples of malware behavior.
- Retrain the ML model using the now balanced dataset (after integrating the GAN generated malware synthetic dataset in the original dataset).
- Test the model on the held-out validation set and submit the output. 
The final evaluation will be carried out by obtaining the GAN trained model RMSE and computing the absolute value of the difference with the baseline model's RMSE. In case of a tie on this difference values, the confusion matrix of the GAN trained model will be considered to asses the quality of malign sample generation.

## Acknowledgment:
This work is jointly supported between the [Center for Cyber Physical Systems (C2PS) at Khalifa University](https://www.ku.ac.ae/c2ps); and the [Technology Innovation Institute (TII)](https://www.tii.ae/) under Fund number 8434000379.