Member Avatar for davy_yg

Can anyone help me explain what is Machine Learning?

I have been googling for the subject and still do not understand what it is. Please help.

1) How could a Machine learn about information? By finding an equation of a pattern? or by having the user inputing a new information about the subject that the machine do not understand?

2) Please gives me a concrete example of Machine Learning.

Thanks in advance.

  • 7 Contributors
  • 6 Replies
  • 540 Views
  • 2 Years Discussion Span
  • Latest Post Latest Post by Reverend Jim

Dani AI

Generated

Contributors like and gave useful analogies, and pointed to neural nets. A compact, practical complement clarifies how a machine actually "learns" and what to do first.

At a technical level, learning means choosing numbers inside a chosen model so its outputs match example data. The usual pipeline is: turn raw inputs into numeric features, pick a model type (simple equation, tree, neural net), define a loss that measures error, and run an optimization that nudges parameters to reduce that loss. Crucial steps that often get left out are splitting data (training/validation/test) and checking that performance on new data matches the training performance (to detect overfitting).

Concrete example — spam detection: collect a set of emails labeled spam/ham, convert text to numeric features (bag-of-words or TF-IDF), split into train/validation/test, fit a simple baseline classifier (e.g., Naive Bayes), and evaluate with precision and recall rather than raw accuracy because spam is often rare. After a reliable baseline exists, iterate on features, try stronger models, and set up monitoring because real data distributions change over time.

Practical checklist and minimal code sketch:

  • Always start with a simple baseline and clear metric.
  • Use cross-validation for model selection.
  • Watch for data leakage and mislabeled examples.
  • Treat feature design as the most effective lever.
from sklearn.model_selection import train_test_split
from sklearn.feature_extraction.text import TfidfVectorizer
from sklearn.naive_bayes import MultinomialNB
from sklearn.metrics import precision_score, recall_score

Xtr, Xte, ytr, yte = train_test_split(emails, labels, test_size=0.2, random_state=42)
vect = TfidfVectorizer(); Xtr_vec = vect.fit_transform(Xtr); Xte_vec = vect.transform(Xte)
m = MultinomialNB(); m.fit(Xtr_vec, ytr); pred = m.predict(Xte_vec)
print("precision:", precision_score(yte, pred), "recall:", recall_score(yte, pred))

Common pitfalls are small/noisy datasets, imbalanced classes, and evaluation on data that leaked from training. Starting small, measuring carefully, and iterating on features gives a clear path from intuition to a working ML system.

Recommended Answers

All 6 Replies

Member Avatar for ddanbe

Machine learning can be all about implementing a

Member Avatar for ryantroop

The best way I had machine learning explained to me was similar to this:

Lets say you want a computer to identify a color. Lets take "Red", or 255, 0, 0 (or some threshold of).
(NOTE: Of course, what you would write is an algorithm that looks at color and a specific threshold (say... "red") not always "red"... because that would just be silly, right?)

You write an algorithm that looks at pixel data, and returns a 1 or a 0: 1 being "YES! It's RED!" and 0 being "NO!" (or Yes! It's a thing, or no it's not!)

You then "train" your algorithm by giving it lots of variations on the color "RED" to include "Like Red" (again, to some threshold of)
You put this training system into separate computers, all of which take an input (your color) and an output (1 or 0)
You then make a program that takes in your color, sends the color randomly to a random number of these computers, and gives you the weighted average of their deduction of if the value is 1 or 0.

You can, then, make adjustments to how to "train" your machines by either letting them figure out variants on their own (by programming that in by giving it the ability to make assumptions, or some memory/database) or by manually saying "You got it right!" or "You got it wrong!" and giving it guided teaching (which it stores those responses instead of making its own assumptions).

While this is super duper simplistic - this is the heart of machine learning. You boil down a physical "thing" to a binary state, and you program machines in however complex a system you like to identify your "thing" (which can be very comlex: for example - a face. How do you identify a face? You have some machines process "eyes", some machines process a "nose", some machines process a "head on a body" some machines process a "mouth" or "teeth" -- and even then, those "machiens" can actually be another series of machines that break down that identification process into smaller digestible 1's and 0's. In the end, you get a weighted average of either a 1 or a 0, and you have the probability that you are looking at a "face." Which is why some machines can be tricked into what it thinks is a face because we give it components, but they may not be in the correct position, color, etc...).

That's about the best I got on it... Hope that helps..

Also, if anyone wants to expand or correct me, I would love to read it.

Ryan

commented: Great info. +0
commented: Really good answer! +0
Member Avatar for digital_ninja

Here's a blog on Understanding how works and how they can change your work and life

Member Avatar for Richard_9

Machine learning is an application of artificial intelligence (AI) that provides systems the ability to automatically learn and improve from experience without being explicitly programmed. Machine learning focuses on the development of computer programs that can access data and use it learn for themselves.
You can read more on AI and ML in this blog : https://blog.eduonix.com/?s=machine+learning

Member Avatar for hussainmujtaba

What is machine learning is a broader topic to cover for this answer.Hope these two explinations help you understand the very basics of ML

  1. Since you mentioned about equation and all , let the first explination be about that. So given some equation of a curve or line,it is common to find the points on that curve, i guess we all have done it in our high school.But suppose given some points ,to be more precise a lot of points, can we find a line or curve that can best fit the points?To be clear by fit, i do not mean pass through all points but be closer to all points like Regression in statistics. Linear regression which is one of the ML algorithms which works just like this.It find a best fit line for all the points.If this is not clear just move to next point
  2. This is much more of an overview of machine learning.Given a lot of samples along with their respective labels, we feed them into our machine learning algorithm and expect the algorithm to correctly label the items it has not seen before.
    Let me give you an example of a child, as most of the children fail to differtiate real food from non eatables like insects, plastic etc.how would you teach him/her to avoid eating insects.One way is that you probably show him a lot of eatables and allow to eat them and scould him/her when approaching to eat insects.This way the child will soon to know what is good to eat and what to avoid eating.
    This is not exactly how machine learning works but i guess you got an idea.
    If you are not satisfied with the explination here, check out this article for a more indetail explination What is machine learninig?
Edited by hussainmujtaba
Be a part of the DaniWeb community

We're a friendly, industry-focused community of developers, IT pros, digital marketers, and technology enthusiasts meeting, networking, learning, and sharing knowledge.