Problem Statement
Lately, an internet network company in Japan has been facing huge losses due to malicious server attacks. They've encountered breach in data security, reduced data transfer speed and intermittent breakdowns in user-user & user-network connections.
When asked, a company official said, " there's a significant dip in the number of active users on our network ".
The company is looking are some predictive analytics solution to help them understand, detect and counter the attacks and make their network connection secure.
Think of a connection as a sequence of TCP packets starting and ending at some well defined times, between which data flows to and from a source IP address to a target IP address under some well defined protocol.
In total, there are 40 type of attacks to which their network is vulnerable to. But, 3 of them cause the maximum damage. In this challenge, you are given an anonymised sample dataset of server connections. You have to predict the type of attack(s).
Download Dataset
Data Description
You are given three files to download: train, test and samplesubmission. The data contains anonymised set of features. Train data has 169307 rows and test data has 91166.
| Variable | Description |
|---|---|
| connection_id | unique id |
| cont_x | numeric features |
| cat_x | categorical features |
| target | target variable (0, 1, 2 are attack types) |
Submission
A participant has to submit a csv file with connection_id and targets as prediction labels. Check the sample submission file for reference. Also, share your source code.
connection_id, target
cxcon_2, 2
cxcon_5, 0
cxcon_8, 0
cxcon_11, 1
cxcon_14, 0
Evaluation Metric
Submission will be evaluated based on multi-classification accuracy metric.
Scripts
- Baseline Models (Python) - Click Here
- Baseline Models (R) - Click Here
Updates
- [Nov 6, 10:55] - Some duplicated IDs in the dataset were found. We have updated the dataset. Kindly download the dataset again and make your submission based on new files. You can continue using your existing models. Sorry for the inconvenience.
- [Nov 6, 11:00] - Submissions based on previous dataset will result in error now.
