MACHINE LEARNING ESCAPE ROOMS
Complete this set of escape rooms, and become a Machine Learning Master!
Start
By : Ashrika Ajanthakumar & Paulina Ferrer
YOUR MISSION
You must escape an industrial facility without being discovered, advancing by answering key ML questions. Use hints when needed. Your challenge starts now!
Industrial Installation
ROOM 1 QUESTION 1
Click here for a hint...
Instructions: Act as a ML model that predicts walking time from home to train station. Review the features, which describe walking conditions. Each feature increases or decreases the base walking duration.
Predict the total travel time based on the new data provided on the left. Give your answer in minutes.
Base walking time
Feature 1: Weather adjustments
Feature 2: School traffic
Feature 3: Time of day
Feature 4: Roadworks
Walking alone: 18 minutes Walking with someone: 23 minutes
Sunny: +0 minRainy: +3 min Snowy: +7 min
Daytime: +0 min Night time: −2 min
At school rush hour: +2 min Not at school rush hour: +0 min
Main roads: +3 minSmaller roads: +1 min No roadworks: +0 min
ROOM 1 QUESTION 2
Industrial Installation
ROOM 2
Click here for a hint...
Instructions: You calculated a travel time of 25 minutes. Based on three categories (Short is <10 min, Medium is 10-20 min, Long is >20 min), a different ML model would predict the outcome as a long walk.
Industrial Installation
ROOM 3
Click here for a hint...
Instructions: These numbers were provided to a ML model for them to be grouped. Nothing else was shared. 3, 4, 5, 7, 8, 12, 13, 14, 15, 17, 22, 23, 24, 40, 41, 42, 60, 86, 88, 89. There were two different approaches to do this.
First approach: the model grouped them like this:Group 1: 3, 4, 5, 7, 8 Group 2: 12, 13, 14, 15, 17 Group 3: 22, 23, 24 Group 4: 40, 41, 42 Group 5: 86, 88, 89 Group 6: 60
Second approach: the model grouped them like this:All loop digits (each digit has a loop): 8, 60, 86, 88, 89 Straight line only (digits from 1/4/7 only): 4, 7, 14, 41, 17 Curvy-no-loop only (digits from 2/3/5 only): 3, 5, 22, 23 Mixed shapes (combos of straight + curvy, no loops): 12, 13, 15, 24, 42 Contains a 0 (loop present): 40
Industrial Installation
Click here for a hint...
ROOM 4
Instructions: Your warehouse AI learns from experience. Whenever it takes an action (like moving, picking, scanning), it gets an outcome. Outcomes can be good (reward) or bad (penalty).
Industrial Installation
ROOm 5
Click here for a hint...
Instructions: ML can work with both structured and unstructured data. Structured data gives the model clearly organised information, while unstructured data requires the model to interpret more complex inputs. Identify whether each example of data source below is structured or unstructured data.
Industrial Installation
ROOM 6
Click here for a hint...
Instructions: Identify which type of model each case study represents:
Model 1
Coca‑Cola uses ML models to analyse product‑interaction data from its beverage dispensers and digital vending machines. The models are trained using labelled data such as customer flavour selections, usage frequency, time‑of‑day patterns, and local purchase trends. The goal is to determine whether a new product variant should be introduced in a specific market or not.
Model 2 An FMCG customer analytics consultant uses ML to simplify a large dataset containing dozens of customer behaviour variables, including purchase frequency, brand preference, promotion responsiveness, and household characteristics. The model compresses this high dimensional dataset into a smaller set of meaningful components, making it easier to visualise and segment customers for further analysis.
Industrial Installation
ROOM 7
Click here for a hint...
Instructions: As per the previous room, identify which type of model each case study represents:
Model 3
Nestlé applies ML forecasting models to improve planning across its global supply chain. These models are trained using labelled data such as historical sales trends, weather conditions, promotional calendars, and supplier performance, enabling the system to estimate future product demand more accurately.
Model 4
Amazon Uses AI systems that learn how to optimise item placement in packing or bin allocation tasks by repeatedly trying different strategies. These systems learn from experience with item sizes, bin capacities, and rewards linked to space utilisation, improving decisions over time.
Model 5
PepsiCo collaborates with retail partners to extract patterns from shopper data and identify groups of stores with similar buying behaviour. The underlying data includes order frequency, category mix, average spend, and customer demographics, which are analysed without predefined labels.
Industrial Installation
ROOM 8
Click here for a hint...
Instructions: In machine learning, we want models that learn patterns that generalise, rather than simply memorising the training data or making random guesses. To assess how well a model is learning, we often refer to the concepts of underfitting and overfitting. You will see several short scenarios describing how a model behaves. Classify each case as overfitting or underfitting.
Industrial Installation
ROOM 9
Click here for a hint...
Instructions: A confusion matrix shows how often a classification model’s predictions match the real outcomes.Using a model that predicts whether a customer will order tomorrow, it compares the predicted label with what actually happened for each case. Identify if the below cases are: True Positive, False Positive, True Negative, or False Negative.
Industrial Installation
ROOM 10 QUESTION 1
Click here for a hint...
Instructions: Decision trees are a common model architecture that can be used for both regression and classification. It is important to distinguish between the task and the architecture: regression is the task, and a decision tree is one type of architecture that can perform that task. Below are two decision trees created to predict transit time under different conditions. Because the prediction is a number, this is a regression task.
Parameters:Base Route = 12 minutes , Traffic = Low , Weather = Good , Stops = 3
Using the parameters, the model predicted 15 minutes. Select which decision tree it must be using:
ROOM 10 EXPLAINED
If you were unsure about room 10, here is the explanation:
This decision tree is performing a regression task (decision trees can also do classification). The model predicted 15 minutes, so the correct architecture belongs to Tree A. You reach that answer by following the highlighted branches according to the given feature values (low traffic, good weather, three stops).
When you reach the end of a branch, called final leaf node, you get the prediction. Decision trees are a foundational model architecture that underpins more advanced ML methods like Random Forests and Gradient Boosting. Click the arrow to proceed.
ROOM 10 QUESTION 2
YOUR FINAL CODE IS:SUCCESS Enter this when prompted to escape! Click here!
YOUR MISSION IS ALMOST COMPLETE. Enter the code from the previous room to escape:
MISSION COMPLETE
You have discovered the truth without being seen; you are the Machine Learning Master! If you want to access a record of the answers, CLICK HERE. Use the code 'SUCCESS' to unlock the file. All capitals - it is case sensitive. You may now close this window and return to Centrical to get your points. Please allow 2 minutes for the points to show on your score.
HINT
Think about whether the information is already organised in rows and columns, or whether the model would need to interpret and extract meaning from the raw content before using it.
Try Again! Use the hint to match each example to the correct type category.
Try Again! Use the hint to match each example to the correct type of data.
HINT
This is an unsupervised learning model that finds patterns or groups in data without labels. A reminder of common ML models: classification, regression, clustering, dimensionality reduction, and reinforcement learning.
WARNING
You are about to enter your final, and most challenging room. Think carefully but don't panic. Use your hint if needed.Click here to proceed....
HINT
Supervised learning includes regression, which predicts a number, and classification, which predicts a category. Unsupervised learning includes clustering, which finds natural groups, and dimensionality reduction, which compresses many features into fewer components. A separate ML approach is reinforcement learning, where a model learns from experience through feedback over time.
HINT
Remember these are some types of models: classification (predicts a category), regression (a number), clustering (find natural groups), dimensionality reduction (compresses features into fewer components), and reinforcement learning (learns from experience through rewards and penalties).
HINT
Follow the branches that match the given parameters and calculate the result for each tree. One tree will produce the exact prediction stated in the problem, while the other will give a similar, but not the exact value. The full explanation will come next.
HINT
Underfitting happens when a model is too simple and fails to learn the important patterns in the data. Overfitting happens when a model learns the training data too closely, including noise, and performs poorly on new data. A good model sits in the middle: not too simple, not too specific.
HINT
The goal of the model has changed, and now the prediction is not a number but a category. Therefore, it is not a regression model anymore. What does it become instead?
Remember these are some types of models: classification (predicts a category), regression (a number), clustering (find natural groups), dimensionality reduction (compresses features into fewer components), and reinforcement learning (learns from experience through rewards and penalties).
THE FINAL CODE WAS
SUCCESS
HINT
The model makes a prediction, and this can be positive or negative. The real outcome determines whether the prediction was true or false. So, if the model predicts that a customer will place an order, that is a positive prediction. But if the customer did not order, then the prediction was false. Therefore, this case is classified as a False Positive.
HINT
For the prediction, start with the base time of walking with someone, then add or subtract the minutes for each condition.
Remember these are some types of models: classification (predicts a category), regression (a number), clustering (find natural groups), dimensionality reduction (compresses features into fewer components), and reinforcement learning (learns from experience through rewards and penalties).
HINT
Supervised learning includes regression, which predicts a number, and classification, which predicts a category. Unsupervised learning includes clustering, which finds natural groups, and dimensionality reduction, which compresses many features into fewer components. A separate ML approach is reinforcement learning, where a model learns from experience through feedback over time.
MACHINE LEARNING V1
COBS
Created on February 4, 2026
Start designing with a free template
Discover more than 1500 professional designs like these:
View
Reboot Protocol
View
Science Breakout
View
Mystery Breakout
View
Musical Room Escape
View
Submarine Escape Game
View
Earth Day Escape Room
View
Corporate Escape Room: Operation Christmas
Explore all templates
Transcript
MACHINE LEARNING ESCAPE ROOMS
Complete this set of escape rooms, and become a Machine Learning Master!
Start
By : Ashrika Ajanthakumar & Paulina Ferrer
YOUR MISSION
You must escape an industrial facility without being discovered, advancing by answering key ML questions. Use hints when needed. Your challenge starts now!
Industrial Installation
ROOM 1 QUESTION 1
Click here for a hint...
Instructions: Act as a ML model that predicts walking time from home to train station. Review the features, which describe walking conditions. Each feature increases or decreases the base walking duration.
Predict the total travel time based on the new data provided on the left. Give your answer in minutes.
Base walking time
Feature 1: Weather adjustments
Feature 2: School traffic
Feature 3: Time of day
Feature 4: Roadworks
Walking alone: 18 minutes Walking with someone: 23 minutes
Sunny: +0 minRainy: +3 min Snowy: +7 min
Daytime: +0 min Night time: −2 min
At school rush hour: +2 min Not at school rush hour: +0 min
Main roads: +3 minSmaller roads: +1 min No roadworks: +0 min
ROOM 1 QUESTION 2
Industrial Installation
ROOM 2
Click here for a hint...
Instructions: You calculated a travel time of 25 minutes. Based on three categories (Short is <10 min, Medium is 10-20 min, Long is >20 min), a different ML model would predict the outcome as a long walk.
Industrial Installation
ROOM 3
Click here for a hint...
Instructions: These numbers were provided to a ML model for them to be grouped. Nothing else was shared. 3, 4, 5, 7, 8, 12, 13, 14, 15, 17, 22, 23, 24, 40, 41, 42, 60, 86, 88, 89. There were two different approaches to do this.
First approach: the model grouped them like this:Group 1: 3, 4, 5, 7, 8 Group 2: 12, 13, 14, 15, 17 Group 3: 22, 23, 24 Group 4: 40, 41, 42 Group 5: 86, 88, 89 Group 6: 60
Second approach: the model grouped them like this:All loop digits (each digit has a loop): 8, 60, 86, 88, 89 Straight line only (digits from 1/4/7 only): 4, 7, 14, 41, 17 Curvy-no-loop only (digits from 2/3/5 only): 3, 5, 22, 23 Mixed shapes (combos of straight + curvy, no loops): 12, 13, 15, 24, 42 Contains a 0 (loop present): 40
Industrial Installation
Click here for a hint...
ROOM 4
Instructions: Your warehouse AI learns from experience. Whenever it takes an action (like moving, picking, scanning), it gets an outcome. Outcomes can be good (reward) or bad (penalty).
Industrial Installation
ROOm 5
Click here for a hint...
Instructions: ML can work with both structured and unstructured data. Structured data gives the model clearly organised information, while unstructured data requires the model to interpret more complex inputs. Identify whether each example of data source below is structured or unstructured data.
Industrial Installation
ROOM 6
Click here for a hint...
Instructions: Identify which type of model each case study represents:
Model 1 Coca‑Cola uses ML models to analyse product‑interaction data from its beverage dispensers and digital vending machines. The models are trained using labelled data such as customer flavour selections, usage frequency, time‑of‑day patterns, and local purchase trends. The goal is to determine whether a new product variant should be introduced in a specific market or not.
Model 2 An FMCG customer analytics consultant uses ML to simplify a large dataset containing dozens of customer behaviour variables, including purchase frequency, brand preference, promotion responsiveness, and household characteristics. The model compresses this high dimensional dataset into a smaller set of meaningful components, making it easier to visualise and segment customers for further analysis.
Industrial Installation
ROOM 7
Click here for a hint...
Instructions: As per the previous room, identify which type of model each case study represents:
Model 3 Nestlé applies ML forecasting models to improve planning across its global supply chain. These models are trained using labelled data such as historical sales trends, weather conditions, promotional calendars, and supplier performance, enabling the system to estimate future product demand more accurately.
Model 4 Amazon Uses AI systems that learn how to optimise item placement in packing or bin allocation tasks by repeatedly trying different strategies. These systems learn from experience with item sizes, bin capacities, and rewards linked to space utilisation, improving decisions over time.
Model 5 PepsiCo collaborates with retail partners to extract patterns from shopper data and identify groups of stores with similar buying behaviour. The underlying data includes order frequency, category mix, average spend, and customer demographics, which are analysed without predefined labels.
Industrial Installation
ROOM 8
Click here for a hint...
Instructions: In machine learning, we want models that learn patterns that generalise, rather than simply memorising the training data or making random guesses. To assess how well a model is learning, we often refer to the concepts of underfitting and overfitting. You will see several short scenarios describing how a model behaves. Classify each case as overfitting or underfitting.
Industrial Installation
ROOM 9
Click here for a hint...
Instructions: A confusion matrix shows how often a classification model’s predictions match the real outcomes.Using a model that predicts whether a customer will order tomorrow, it compares the predicted label with what actually happened for each case. Identify if the below cases are: True Positive, False Positive, True Negative, or False Negative.
Industrial Installation
ROOM 10 QUESTION 1
Click here for a hint...
Instructions: Decision trees are a common model architecture that can be used for both regression and classification. It is important to distinguish between the task and the architecture: regression is the task, and a decision tree is one type of architecture that can perform that task. Below are two decision trees created to predict transit time under different conditions. Because the prediction is a number, this is a regression task.
Parameters:Base Route = 12 minutes , Traffic = Low , Weather = Good , Stops = 3
Using the parameters, the model predicted 15 minutes. Select which decision tree it must be using:
ROOM 10 EXPLAINED
If you were unsure about room 10, here is the explanation:
This decision tree is performing a regression task (decision trees can also do classification). The model predicted 15 minutes, so the correct architecture belongs to Tree A. You reach that answer by following the highlighted branches according to the given feature values (low traffic, good weather, three stops).
When you reach the end of a branch, called final leaf node, you get the prediction. Decision trees are a foundational model architecture that underpins more advanced ML methods like Random Forests and Gradient Boosting. Click the arrow to proceed.
ROOM 10 QUESTION 2
YOUR FINAL CODE IS:SUCCESS Enter this when prompted to escape! Click here!
YOUR MISSION IS ALMOST COMPLETE. Enter the code from the previous room to escape:
MISSION COMPLETE
You have discovered the truth without being seen; you are the Machine Learning Master! If you want to access a record of the answers, CLICK HERE. Use the code 'SUCCESS' to unlock the file. All capitals - it is case sensitive. You may now close this window and return to Centrical to get your points. Please allow 2 minutes for the points to show on your score.
HINT
Think about whether the information is already organised in rows and columns, or whether the model would need to interpret and extract meaning from the raw content before using it.
Try Again! Use the hint to match each example to the correct type category.
Try Again! Use the hint to match each example to the correct type of data.
HINT
This is an unsupervised learning model that finds patterns or groups in data without labels. A reminder of common ML models: classification, regression, clustering, dimensionality reduction, and reinforcement learning.
WARNING
You are about to enter your final, and most challenging room. Think carefully but don't panic. Use your hint if needed.Click here to proceed....
HINT
Supervised learning includes regression, which predicts a number, and classification, which predicts a category. Unsupervised learning includes clustering, which finds natural groups, and dimensionality reduction, which compresses many features into fewer components. A separate ML approach is reinforcement learning, where a model learns from experience through feedback over time.
HINT
Remember these are some types of models: classification (predicts a category), regression (a number), clustering (find natural groups), dimensionality reduction (compresses features into fewer components), and reinforcement learning (learns from experience through rewards and penalties).
HINT
Follow the branches that match the given parameters and calculate the result for each tree. One tree will produce the exact prediction stated in the problem, while the other will give a similar, but not the exact value. The full explanation will come next.
HINT
Underfitting happens when a model is too simple and fails to learn the important patterns in the data. Overfitting happens when a model learns the training data too closely, including noise, and performs poorly on new data. A good model sits in the middle: not too simple, not too specific.
HINT
The goal of the model has changed, and now the prediction is not a number but a category. Therefore, it is not a regression model anymore. What does it become instead? Remember these are some types of models: classification (predicts a category), regression (a number), clustering (find natural groups), dimensionality reduction (compresses features into fewer components), and reinforcement learning (learns from experience through rewards and penalties).
THE FINAL CODE WAS
SUCCESS
HINT
The model makes a prediction, and this can be positive or negative. The real outcome determines whether the prediction was true or false. So, if the model predicts that a customer will place an order, that is a positive prediction. But if the customer did not order, then the prediction was false. Therefore, this case is classified as a False Positive.
HINT
For the prediction, start with the base time of walking with someone, then add or subtract the minutes for each condition. Remember these are some types of models: classification (predicts a category), regression (a number), clustering (find natural groups), dimensionality reduction (compresses features into fewer components), and reinforcement learning (learns from experience through rewards and penalties).
HINT
Supervised learning includes regression, which predicts a number, and classification, which predicts a category. Unsupervised learning includes clustering, which finds natural groups, and dimensionality reduction, which compresses many features into fewer components. A separate ML approach is reinforcement learning, where a model learns from experience through feedback over time.