Making Sense of Data: Unpacking Machine Learning Basics


Machine learning (ML) unlocks the potential of data by enabling computers to learn and perform tasks without explicit programming. Think of it as teaching computers through data, enabling them to make decisions or predictions. This article unpacks the basic concepts of ML.

Data as Building Blocks

Think of countless data points as the building blocks of ML. Each holds a piece of information, be it a customer’s purchase history, a medical record, or website clicks. These diverse points come in various forms:

  • Numbers: Quantifiable data like prices, clicks, or temperatures.
  • Categories: Discrete options like male/female, geographic regions, or spam/not spam flags.
  • Text: Product reviews, website searches, or social media posts.
  • Multimedia: Images, audio recordings, or music files.

Structured vs. Unstructured Data

Data can be organized like a tidy spreadsheet (structured) or resemble a pile of documents and photos (unstructured). This distinction influences how ML approaches learning, requiring different techniques for each type.

Machine Learning vs. Data Science

While machine learning plays a critical role in extracting insights from data, it’s important to distinguish it from data science.

Data science is the science of extracting knowledge and insight from data. It encompasses the entire data analysis process, from data collection and cleansing to analysis, communication, and action. It uses a variety of techniques and tools, including machine learning, statistics, data visualization, and domain knowledge. The output of a data science project is often a set of insights that can help an organization make business decisions. Data science is a subset of artificial intelligence.

Machine learning is a subfield of artificial intelligence and also partially a subfield of data science. ML focuses on building algorithms that can learn from data. ML models learn the patterns, the basic structure in the data used to train them. Based on that training, the models learn to make predictions or decisions when confromnted with new data.

ML is a set of tools and concepts used in data science, but it has also transformed many fields outside of data science. To name just a few examples, the impact of ML can be felt in the game of chess, the speed and accuracy of medical diagnoses, the efficiency of financial transactions, and the personalization of online experiences. An ML project often results in a piece of software.


Data science vs Machine learning and Artificial intelligence

Image source: www.collegevidya.com/blog/data-science-vs-artificial-intelligence-vs-machine-learning/

The Two Main Machine Learning Tasks

ML tackles two primary tasks:

  1. Classification: Assigning labels or categories to data points. For example, an ML algorithm trained on images can accurately label them as cats or dogs. Similarly, it can analyze text sentiment, classifying tweets or reviews as positive, negative, or neutral.
  2. Quantitative Prediction (regression): Estimating numerical values. Imagine an ML algorithm that predicts temperature and precipitation based on historical data and current conditions. It can also use past purchases or browsing history to recommend products or services a customer might be interested in.

Choosing the Right Tool for the Job: Selecting the Optimal Machine Learning Model

So you have your data and understand its basic structure. Now comes the critical step: choosing the right ML model for your specific task. While experimentation plays a role, a strategic approach leads to a more informed and efficient choice:

1. Understand your problem: Clearly define your prediction or classification goal. Are you aiming for binary classification (e.g. spam/not spam), multi-class (e.g. image recognition), or numerical prediction (e.g. temperature)? This guides your search towards relevant model types.

2. Consider data properties: Analyze the size, complexity, and structure of your data. Large datasets may favor scalable models such as support vector machines (SVMs) or random forests, while smaller datasets may work well with decision trees or linear regression. Structured data often suits rule-based models, while unstructured data may require deep learning approaches.

3. Leverage domain knowledge: Your domain expertise is often invaluable. Are there established models that are commonly used for similar problems? Can you identify features that affect the prediction? This knowledge helps you eliminate inappropriate models and fine-tune the ones you choose.

4. Explore Benchmarking and Performance Metrics

4.1. Benchmarking

Suppose you’ve trained a model to recognize handwritten digits. To benchmark its performance, you can test it on the MNIST dataset, a widely used standard dataset for handwritten digit recognition. By comparing your model’s accuracy on this dataset with other published results on the same dataset, you can evaluate its relative performance against other algorithms performing the same task.

Imagine you’ve developed a model to predict which customers are likely to churn (stop using your service). You could benchmark its performance by comparing its churn prediction accuracy to industry averages or to the performance of similar models used by competitors.

In summary, benchmarking provides an external reference point for understanding your model’s performance. It can be based on established datasets, industry standards, or competitor performance. It helps you assess the strengths and weaknesses of your model in a broader context.

4.2. Performance Metrics

In ML, performance metrics are the score card for your model. Just like checking the mileage on your car, these metrics tell you how well it performs on your specific data. Different tasks require different “scores”. For guessing images (classification), accuracy shows overall success, while precision shows how many correct guesses are actually true. For predicting numbers (regression), Mean Squared Error measures how close your guesses are to reality.

5. Experiment with focus: After narrowing down the options, conduct focused experiments to compare and refine a few promising models. Use appropriate evaluation metrics that are relevant to your task.

Image source: www.mapendo.co/blog/training-data-the-milestone-of-machine-learning


Cite this blog post
Stéphane Condorelli (2024, February 2). Making Sense of Data: Unpacking Machine Learning Basics. Exploring Large Language Models. Retrieved June 22, 2024, from https://doi.org/10.58079/w76k

Leave a comment

Your email address will not be published. Required fields are marked *

This site uses Akismet to reduce spam. Learn how your comment data is processed.

Search OpenEdition Search

You will be redirected to OpenEdition Search