GST Analytics Hackathon 2024 – Cash Prize Upto Rs. 25 Lakhs !

Structure of the Hackathon:

• The Hackathon would be organised as an online event with processes for registering participants, accessing the datasets for each problem statement, and submitting developed prototypes. There would be an offline event with the shortlisted participants for the finale/second round.
• Indian students, researchers from educational institutions, or professionals working in Indian startups and companies can participate in the Hackathon. The participant must be a citizen of India.
• The participants are expected to form teams of up to five members, including at least one team lead. A participant may only register as a member of a single team.
• The Hackathon would take place over 45 days from the start of registration to the final date for submission of developed prototypes.
• Participants would receive a dataset containing 9 lakh records with around 21 attributes each. The data is anonymized and labelled, including trained, validated, and non-validated datasets.
• Before submitting the solution prototype, participants must upload their code in the GIT (https://www.github.com) repository and an optional demo/product video on YouTube.
• For online submissions, the following required/optional fields are to be shared for evaluation:
  • Idea/Concept
  • Project Description
  • Source Code URL (github.com)
  • Video URL
  • GitHub Unique Source Code Checksum – Steps to create checksum are mentioned in later steps.
  • Project Report

• The evaluation process of the Hackathon would be overseen by a distinguished panel of jury members comprising experts from the fields of machine learning, data science, and tax administration. The jury would rigorously assess each submission based on predefined criteria to ensure a fair and comprehensive evaluation.

Prizes:

The Hackathon offers significant prizes for the top-performing teams, and these are:

• First Prize: Rs. 25 lakhs
• Second Prize: Rs. 12 lakhs
• Third Prize: Rs. 7 lakhs
• Special Prize of Rs. 5 lakhs for All-Women Teams (in addition to the top three prizes)
• Prizes will only be awarded if the model created meets the jury’s satisfaction with the usability of the designed solution as a viable product.
• Consolation prizes of Rs. 3 lakh, Rs. 2 lakh, Rs. 1.5 lakh and Rs. 1 lakh would be given instead of announced prizes if the jury does not find any model to solve the problem statement.

The evaluation process of the Hackathon would be overseen by a distinguished panel of jury members comprising experts from machine learning, data science, and tax administration. The jury would rigorously assess each submission based on predefined criteria to ensure a fair and comprehensive evaluation.

Jury Composition: The jury would tentatively include:

• Senior data scientists with extensive experience in predictive modelling and AI.
• Tax administration experts with a deep understanding of fraud detection and related challenges.
• Academic professionals specializing in machine learning and data analytics.
• Representatives from GSTN and NIC with domain-specific expertise.

Evaluation Process:

• Initial Screening: Submissions would undergo an initial screening to ensure compliance with submission guidelines and basic functionality.
• Technical Evaluation: The jury would conduct a detailed technical evaluation of the models with the help of GSTN’s data team, focusing on performance metrics, innovation in approach, and robustness of the solution.
• Practical Usability: Models would be assessed for their practical usability and potential for real-world implementation
• Based on the initial evaluation, 9 to 15 teams would be shortlisted for the second round.
• In the second round, teams would refine their models using additional data and insights from discussions with SMEs. The final submissions would include a fine-tuned model, a detailed write-up, and a presentation, followed by an interview with the jury in Delhi.

Decision Making:

• Prizes would be awarded to the top three teams whose models meet the jury’s satisfaction regarding usability as viable products. A special prize would be awarded to the women-only team, if any.
• If no solution meets the required standards, consolation prizes would be awarded based on the jury’s discretion.
• The jury’s decision would be final and binding.

• The efficiency and effectiveness of the proposed algorithms will be evaluated against the validation dataset. This rigorous assessment would determine the models’ practical viability and accuracy in real-world applications.
• Submitted models (in first and final rounds) would be evaluated on the following metrics, given that we have provided a binary classification problem. Participants are encouraged to provide the following metrics along with the algorithm (model) at the time of final submission:
  • Accuracy: The proportion of correctly classified instances (both true positives and true negatives) out of the total instances
  • Precision: The proportion of true positive instances out of those predicted as positive.
  • Recall (Sensitivity or True Positive Rate): The proportion of true positive instances out of the actual positive instances.
  • F1 Score: The harmonic means of precision and recall provides a metric that balances both concerns.
  • AUC-ROC (Area Under the Receiver Operating Characteristic Curve): AUC represents the degree of separability and measures how well the model distinguishes between classes. ROC plots the true positive rate (Recall) against the false positive rate (1- Specificity).
  • Confusion Matrix: A table that provides a detailed breakdown of true positives (TP), true negatives (TN), false positives (FP), and false negatives (FN). It helps in visualizing the performance of the classification model.
  • Other Metrics (Optional): Log Loss and Balanced Accuracy of the model.
  • Any other additional criteria as decided by jury member.