Coding Club

IIT Guwahati
Coding Week

Machine
Learning
Task Released Deadline: 27th Dec

Exclusively for 1st-year students.*

Machine Learning Task 1

The World Tour of '99: The Data Detective

The Scenario

You are the newly hired data scientist for Electric Omen, a legendary rock band planning
their massive comeback tour. The band has played hundreds of shows across four
legendary venues, and now they need your expertise to ensure every future show is a
riotous success.

Your mission: predict the Crowd Energy (a score from 0-100) for future venues. The band's
manager believes that understanding what drives crowd energy will help them optimize
setlists, pricing, and scheduling for maximum impact.

You have access to the band's historical tour logs (tour_logs_train.csv). However, fair
warning, the band's roadies weren't exactly meticulous record-keepers. The data is chaotic,
messy, and inconsistent.

You also have access to the Lead Singer's personal notes (lead_singer_scribbles.txt). These
cryptic scribbles contain the singer's theories and observations about what affects crowd
energy. IMPORTANT: The singer admits these notes are unreliable, contradictory, and
possibly wrong. Treat them as hypotheses to test, not facts to assume.

Dataset

The below link contains the training and the test datasets, ensure that the test set is in no
way used for the training of your model, doing so will lead to disqualification. (link)

The Four Venues

Each venue has its own personality, quirks, and hidden rules. Part of your job is figuring out
what makes each one tick—and whether the singer's theories hold up.

The Holy Grounds (V_Alpha) - A converted monastery. The singer has theories about
noise limits here.

The Vampire’s Den (V_Beta) - A gothic nightclub. The singer believes timing matters, but
isn't sure how.

The Snob Pit (V_Gamma) - An exclusive venue. The singer suspects pricing plays a role,
but can't remember the details.

The Mosh Pit (V_Delta) - Chaotic crowds. The singer has conflicting memories about
what drives energy here.

Machine Learning Task 1

Y our M ission

Primary Objectives

Data Cleaning & Wrangling: The data is messy. Dates are inconsistent. Prices have mixed
currencies. Some readings are clearly errors. Your first job is to make sense of the chaos.

Exploratory Data Analysis (EDA): Use visualizations to discover patterns. Test the singer's
hypotheses against the actual data—some may be right, some may be wrong, some
may be partially true.

Feature Engineering: Create new features that capture the patterns you've discovered.

Think about interactions, thresholds, and transformations.

Model Training & Hyperparameter Tuning: Train a regression model to predict

Prediction Submission: Generate predictions for the test set in the required format.

Hyperparameter Tuning Requirement

This is a mandatory requirement, not optional. Your submission must include:

Justification for your model choice

Documentation of hyperparameters explored (ranges, values tested)

Validation strategy used during tuning (e.g., k-fold cross-validation)

Final hyperparameter values with reasoning for selection

Comparison of tuned model vs. default parameters

Bonus Objective: Revenue Optimization

The band's manager, Rick, wants to maximize profit at V_Gamma (The Snob Pit). He's given
you the following business context:

Venue capacity: approximately 800 seats

Fixed costs per show: ~$5,000 (venue rental, crew, equipment)

Variable cost per attendee: ~$8 (security, cleaning, insurance)

Rick has noticed that crowd energy affects whether people stay, buy drinks, and buy
merch

Attendance may be affected by both ticket price and crowd energy—but Rick doesn't
know exactly how

Rick's request: "Find me the ticket price that makes the most money. I don't care how you
figure it out—just show me the math and justify it. If you tell me 'it depends on the moon'
I'm docking your pay."

Your task:

Derive your own revenue/profit optimization formula — we are NOT giving you the
formula

State your assumptions explicitly

Use your trained model to simulate different price points

Create a visualization showing your optimization curve

Identify and justify the optimal price point

Module Name Task

Warnings
Read these carefully. Ignoring them will result in failure.

The Scribbles Are UNRELIABLE

The Lead Singer's Scribbles contain theories, guesses, and half-remembered observations.
Some are correct. Some are wrong. Some are partially true. Do NOT treat them as ground
truth. Use them to generate hypotheses, then test those hypotheses against the actual data.

Data Leakage

Some columns in the dataset are collected AFTER the show. Using them to predict
Crowd_Energy is cheating and will result in models that fail catastrophically in production.
Think carefully about what information would actually be available BEFORE a show.

Red Herrings

Not every column matters. The singer has theories about many features, some are noise. A
good data scientist knows what to keep and what to discard based on evidence, not intuition.

Data Quality Issues

Dates are logged in multiple inconsistent formats (including text descriptions)

Prices include different currencies: $, £, €. Exchange rates: £1 ≈ $1.27, €1 ≈ $1.09

Some sensor readings have systematic failures (missing or zero values)

Outliers exist from data entry errors (impossible values)

Distribution Shift

The test set may not have the same distribution as the training set. Models that overfit to
training patterns will fail. Build robust models that generalize.

Unseen Categories

The test set may contain categorical values that don't exist in training. Make sure your
pipeline handles unknown categories gracefully without crashing

Deliverables
[Link] - Your predictions in the required format ( link)

analysis_notebook.ipynb - Complete analysis including EDA, cleaning, feature

findings_report.pdf - 1 to 2 summary highlighting the key findings for each venue, which
of the singer's theories were correct/incorrect, model choice justification

(Bonus) revenue_optimization.pdf - Your formula derivation, assumptions, optimization

Machine Learning Task 1

Evaluation Criteria

You will be evaluated on:

Criterion Weightage

Prediction Accuracy (RMSE on test set) 20

Data Cleaning Quality 15

EDA & Pattern Discovery 20

Feature Engineering 25

Hyperparameter Tuning 10

Code Quality & Documentation 10

Bonus: Revenue Optimization (with formula derivation) +15

Note

Using data leakage features (features that would not be available before the show)

Submitting predictions in incorrect format

Hardcoding test set answers

No evidence of hyperparameter tuning (using only default parameters)

Plagiarism or collaboration (this is an individual assessment)

This challenge is designed to test real data science skills, not just model fitting. We want :

Critical thinking: Can you evaluate the singer's claims against evidence?

Data intuition: Can you recognize leakage, noise, and real signals?

Technical rigor: Can you properly tune and validate a model?

Communication: Can you explain your findings clearly?

Business sense: (Bonus) Can you translate a model into business recommendations?

Good luck. Rock on.

Deadline - 28th Dec

Submission Link

Join the WhatsApp Group

Having any doubts ? Contact us

Manthan : 9512529995

Nilay : 9340251202

Pushpendra : 7351816883

Arnav : 7982390939

Naveen : 8074523143

Sarvesh : 7262006199