26 Datasets For Your Data Science Projects

A compilation of task-based datasets that you can use for building your next data science project.

Photo by Alex Azabache on Unsplash

Looking at Kaggle or Google Datasets, I always find it hard to settle on a dataset to try out a new machine learning concept that I recently learned. Fresh datasets are posted everyday on these popular websites and the effort to find the right one for a new project quickly becomes overwhelming.

With this mind, and because I could not find a single proper task-based compiled list anywhere on the internet, I decided to make one myself — a master list actually, one that I’ll always come back to when I need to choose a dataset to practise a newly acquired skill.

And hopefully, it will be beneficial for you as well!

What this article is ABOUT and NOT ABOUT

Let this article be a reference for you to consider trying out a dataset for a new algorithm that you learn and want to experiment with.

However, please do not consider it to be an exhaustive list of datasets for machine learning tasks — it is not possible for me or anyone to look over every single dataset available on every website out there.

Okay, now that we’ve gotten that over with, let’s get on with the list. 😃

The List

I will try to cover as many machine learning/deep learning-based tasks and related datasets for them as I can think of, and I will also update this article from time to time. So consider bookmarking it for safekeeping as well, if you want.

I’m including the datasets I’ve been using throughout my data science journey — personal favourites included — and also some that I am hoping to try in the future.

The Multi-Purpose Datasets — For trying out any big and small algorithm

1. Kaggle Titanic Survival Prediction Competition — A dataset for trying out all kinds of basic + advanced ML algorithms for binary classification, and also try performing extensive Feature Engineering.

2. Fashion MNIST — A dataset for performing multi-class image classification tasks based on different categories such as apparels, shoes, handbags, etc.

3. Credit Card Approval — A binary classification task for good or bad credit scores, if the people can be a risk for defaulting credit card loans.

4. Rock Paper Scissors— Image classification for those three classes.

Regression Datasets

1. Boston House Prices — A classic dataset for flexing your Regression muscles.

2. WHO Life Expectancy — Good one for experimenting with your EDA skills also.

3. California Housing Prices — almost the same as no. 1.

General Classification Problems

1. Heart Disease UCI — To predict the presence of heart disease in the patient based on a number of features.

2. Heart Attack Analysis — Classify chance of heart attack — for binary classification.

3. Campus Recruitment — Determine if a student gets placed based on various features.

Image Classification Problems

1. Pneumonia Data — For classifying the type of Pneumonia in patient X-Rays.

2. Face Mask Classification —Multi class classification into three types of categories —with mask, without mask, and mask worn incorrectly.

3. Intel Image Classification— A good one for classifying natural sceneries into 6 categories.

Recommender Systems

1. MovieLens Dataset — A classic one for starting out, predict which movie to recommend.

2. Goodreads Books — Detailed information about books through numerous columns for building a book recommender engine. This is my personal favourite for getting a hang out of actually attempting the recommendation task.

3. Netflix Data — collection of movies + TV shows details until 2019, also a great one for some practical exposure to a real world application.

Time Series Analysis

1. E-Commerse Sales — For predicting sales/transaction for a store. The classic time series forecasting job.

2. Minimum Daily Temperatures — This dataset describes the minimum daily temperatures over 10 years in the city Melbourne, Australia.

3. Household Electric Power Consumption — Has measurements of electric power consumption in one household with a one-minute sampling rate over a period of 4 years.

Large Scale Text Datasets for NLP tasks

1. Amazon Reviews — A classic dataset for sentiment analysis task. Overused a little yes, but undeniably a great one.

2. Covid-19 Open Research Challenge — With great number of Covid research articles with full text, this is a great dataset to start out with text summarisation, semantic search and question answering systems.

3. Arxiv Dataset — Collection of arxiv research papers for building text generation, abstractive summarisation, and question answering systems.

4. News Headlines — To detect sarcasm in news headlines, a classification task.

5. StackOverflow Questions — 60k questions on SO to predict tags for questions, as well as classify a question based on scores as high or low quality.

Text — Optical Character Recognition

1. Handwriting Recognition — This contains about four hundred thousand handwritten names, a good dataset for an OCR task.

2. Captcha Recognition

Thanks for reading! I will try to make sure to update this list often, and hopefully also include several more problem types.

I write about Data science on a regular basis and try to make it fun for my readers. Follow me to get new articles straight in your feed. :)

Here is the codebase of all my Data Science stories. Happy learning! ⭐️

Another article of mine that you might wanna read: Making Your First Kaggle Submission An easy-to-understand guide to getting started with competitions and successfully modelling and making your first…towardsdatascience.com