Logistic Regression

Описание: Logistic regression
Have you ever checked the weather and decided whether to bring an umbrella? Maybe if there’s a high chance of rain, you take one with you, but if the forecast looks clear, you don’t bother. Machine learning uses logistic regression to aid in making decisions of this type, which are typically yes or no.
Logistic regression is a classical machine learning model that helps solve problems where the goal is to estimate one of two possible outcomes based on input variables.
For example, a bank might use logistic regression to estimate the likelihood that someone will repay a loan. The model does not directly say “yes” or “no”. Instead, it gives a probability, such as a 70% chance that the person will repay the loan. That number is then used to help make a decision.
While decision trees use step-by-step questions to reach a result, logistic regression uses a curve to calculate the probability based on patterns in the data. Unlike linear regression, which provides insights into values such as price or time, logistic regression estimates the likelihood of an outcome.
Select each section to learn more about logistic regression.
Odds and probability
In logistic regression, the model does not guess yes or no directly. Instead, it first calculates something called odds and probability, then turns those odds into a probability between 0 and 1.
• Odds compare the chance that something will happen to the chance it won’t. For example, if 8 out of 10 people respond to a survey, the odds are 8 to 2 (meaning 8 responded, 2 did not).
• Probability is the chance that something will happen. It’s a number between 0 and 1. In the example, the probability is 8 divided by 10, or 0.8. A 0.8 probability means there is an 80% chance of something happening.
Understanding odds and probability helps explain how logistic regression makes its recommendations.
The logistic (sigmoid) function
The logistic, or sigmoid, function is a mathematical formula that turns odds into a probability value between 0 and 1. The function creates an S-shaped curve, starting low, rising quickly in the middle, and flattening near the top. The shape of the curve means the model is most confident when the number is very high or very low, and least confident when it’s in the middle.
Threshold and binary decision
After logistic regression transforms its results into a probability using the sigmoid function, it still needs to make a final binary decision: Is the answer “yes” or “no”?
That’s where the threshold comes in. A threshold is a set point that determines how a probability is converted into a final yes or no outcome.
In most cases, the threshold is set at 0.5. That means:
• If the probability is above 0.5, the answer is yes.
• If the probability is below 0.5, the answer is no.
This is how logistic regression transforms the odds into a final yes-or-no decision.
Adjusting the threshold alters the model’s sensitivity. A lower threshold means more yes answers, but it also might mean more mistakes. A higher threshold makes the model stricter and less likely to return a “yes” response.
For example, a company might use logistic regression to decide which customers are likely to buy a product. If the model suggests a 70% chance for one customer and a 40% chance for another, the company might set a threshold (such as 0.6) to decide who receives a special offer.
Classical machine learning uses various models to train AI systems to assist in decision making, provide insights, and offer recommendations. Decision trees, linear regression, and logistic regression each approach data and choices in different ways. Next, you’ll use your knowledge of these models to make recommendations within a realistic setting.