DataByte

The Official Machine Learning and Data Science Club of

NITT
FIRST-YEAR INDUCTION

TASK-1

MACHINE LEARNING

All the machine learning models are to be built from scratch without taking the help of
sklearn or any similar library. Otherwise, the submission would not be considered. All the
problem statements involve very basic steps to approach them, so we expect Innovation and
creativity in approach, more than implementing a model.

General Instructions for Submission:

● The submission must include all the EDA performed and the results from them.
● There must be at least 3 very different models implemented and results verified (say
decision trees, KNN, Logistic regression)
● Apart from the given evaluation metric, explore others and suggest why F1 or any
other metric is better for the problem statement you are working on
● Comment on the code correctly and make sure they are readable, and highlight
wherever you have tried something creative or used an innovative approach.
● All the labels have been given for the problem statements given below, but what do
you do if you just have a bunch of numbers without labels? Try implementing that
along with your task for exploring your creativity.

1
Problem Statements:

(Choose any one problem statement)

1. Liver Disease Prediction (Classification):

The task is to develop prediction models utilising 18 clinical characteristics to

forecast the stage of liver cirrhosis. Cirrhosis causes liver harm for a number of
reasons, which results in scarring and liver failure. These types of applications are
very crucial to use AI in healthcare.

This is basically a classification problem where the goal is to categorise the stage of
the disease so that measures can be taken according to it. Before modelling, use
suitable processing techniques and present visually the inference you get from the
EDA. Then proceed to develop the prediction models and compare them using
evaluation metrics, and justify the results statistically to support your model’s
performance.

About data:
Train Dataset - It consists of a total of 6801 data points.
Test Dataset - You must predict the stage of cirrhosis of 3201 data points.

Evaluation: F1 Score

Dataset: Liver_disease_pred

(or)

2. Taxi Fare Prediction (Regression):

Have you ever felt taxis or autos just get you fooled at times, asking for an insanely
high amount, now you will have to approach this as a budding machine learning
enthusiast. Your task is to delve into the world of transportation economics and build
a regression-based model to predict the fare using features like the trip duration,
distance travelled, etc. Problems like this are solved better using domain knowledge
so feel free to take an intuitive approach to this mystery.

About data:
Train: 20967 x 8
Test: 89861, 8

2
Evaluation Metric: Choose the best metric to evaluate for the dataset with statistical reason

Dataset: Fare_prediction

(or)

3. Decode AQI (Classification):

The Air Quality Index (AQI) is a widely-used measure of air pollution that provides
information on the quality of air in a city on a daily basis. The task is to develop a
model which can predict the AQI_Bucket based on features like PM2.5, CO, etc.
This dataset contains air quality measurements for several cities, with observations
taken at different dates and times. The variables represent different pollutants and
their concentrations in the air, as well as the AQI bucket, which provides an overall
measure of air quality."

About data:
Train: 495512 x 15
Test: 212363 x 14

Evaluation: F1 Score

Dataset : AQI
(or)

4. Wine Quality Prediction (Classification):

Embark on a journey into the world of wine, where predicting quality takes an
unexpected twist. Your mission is to develop a model that accurately forecasts wine
quality using freely available datasets from the internet. But here's the catch: the
dataset you'll be working with lacks clear information on the usual factors used to
assess wine quality.

As a curious data scientist, your task is to uncover hidden connections within this
mysterious dataset. By harnessing the power of machine learning, you'll need to
uncover insights that could hold the key to predicting wine quality. Dive deep into
the unknown, examining hidden variables and patterns that might reveal the secrets
of excellence. Your model must adapt and discover the unspoken features that truly
influence wine quality, exploring new territory in the world of data-driven wine
assessment.

3
 Dataset: Wine Quality - UCI Machine Learning Repository

(or)
5. Smart Botanist (Classification):
As a botanist, the task at hand involves developing a machine learning model to
effectively classify different species of flowers. This model will utilise available
features from the dataset to accurately determine the species of a given flower. The
ultimate objective is to create a reliable and robust classification model that can
generalise well to new samples of iris flowers. Take into consideration the effects of
seasonal and climatic changes i.e environmental conditions.(Hint> Soil types, rainfall
etc). Various geographic locations differ in the types of iris flowers present, the
objective is to provide accurate predictions for their species(Do research into
incorporating these points into your dataset while building your model).

Now, consider that the model is able to perform well on the training data but does
not perform well while testing on unseen input features of flowers. What will you do
in order to prevent this? Furthermore, the model's performance will be assessed using
appropriate evaluation metrics such as precision, recall, and F1 score, ensuring a
comprehensive understanding of its effectiveness across various classes. Suggest a
way you can automate the learning process of the model.
Dataset: Iris - UCI Machine Learning Repository

(or)

6. Real Estate Price Prediction (Regression):

Melbourne, being the evergreen city in the beautiful country of Australia, has a
variety of households for the average real estate buyer to choose from. Your
objective is to correctly predict the price of a respective real estate property in
Melbourne, taking into consideration the various factors such as land area,
the number of bedrooms, etc. The model must utilise all factors that may affect the
price of a house in Melbourne with the help of regression techniques, train and test
your dataset so as to evaluate the model with regressive metrics such as (MSE)
Mean Squared Error, RMSE, etc.
Your dataset may have outliers, so do the needful and make sure it can adapt and
improvise to correctly predict the price of a house for a prospective buyer. Be sure to
produce realistic predictions.

Brownie points will be given to those who can do an appropriate EDA on the dataset
and derive meaningful conclusions.

4
 Dataset: House_Pricing

Brownie points:
Devise a deep learning algorithm to improvise the accuracy scores of above-implemented
models developed using machine learning techniques.
(NOTE: This task is optional. But it is highly recommended to do so. Implementing this task
will increase your chance of selection into the club)
Also,
● Try implementing these algorithms from scratch without using libraries
● To further enhance your skills and exposure, try executing the clustering
classification and regression algorithms in different environments. This will allow
you to explore various features and libraries simultaneously.

Expectations:
● Implement it with the help of PyTorch (preferred) or Tensorflow for the task.
● Use Jupyter Notebooks as the coding environment.
● Consider setting up Jupyter in VS Code to execute the algorithms.

Guidelines:
● The read.me file of Github must include the accuracies of algorithms by adding
suitable evaluation metrics scores, graphs of epochs, loss, and accuracy.
● Utilize various evaluation methods such as confusion matrix, ROC curve, or
precision-recall curve to assess the performance of each approach.
● Provide insights into the strengths and limitations of each method based on the
evaluation results.

Github link submission:

Include your code implementation, evaluation metric scores, graphs, and report. (NOTE:
Failing to include any of these will abstain evaluation of your work)

5
Resources:

Setting up Google Colab Notebooks

● Google Colab Tutorial for Beginners | Get Started with Google Colab
Setting up Jupyter Notebooks
● Install Miniconda (Python) with Jupyter Notebook and Setting Up Virtual Environments on
Windows 10
Setting up VS Code
● VSCode Tutorial For Beginners - Getting Started With VSCode
Python
● https://www.python.org/
Virtual Environment
● Create your own python virtual environment using the link below
● https://www.youtube.com/watch?v=ohlRbcasPAc

Machine Learning
● Machine Learning Tutorial Python | Machine Learning For Beginners - YouTube
● https://trainings.internshala.com/machine-learning-
course/?tracking_source=trainings-search-tags
● Practical Machine Learning Tutorial with Python Intro p.1
● https://www.coursera.org/learn/machine-learning or refer to Andrew NG videos on
youtube with the link below#1 Machine Learning Specialization [Course 1, Week 1,
Lesson 1]
● https://www.kaggle.com/learn/intro-to-machine-learning for extensive notes on
various concepts on the basics of ML

Deep Learning
● Go through @patloeber , his courses for deep learning techniques and machine
learning from scratch.
● Deep Learning playlist overview & Machine Learning intro
● PyTorch Prerequisites - Syllabus for Neural Network Programming Course
● https://youtube.com/playlist?list=PLhhyoLH6IjfxVOdVC1P1L5z5azs0XjMsb
● Pytorch
Learn basics of Pytorch and its implementation using the link below:
https://pytorch.org/tutorials/index.html#

Github tutorials
● GitHub Tutorial - Beginner's Training Guide