Continuing our exploration of common distributions, we turn to another important discrete distribution: the Poisson distribution. While the Binomial distribution models the number of successes in a fixed number of trials, the Poisson distribution models the number of times an event occurs within a specified interval of time or space. Think about counts: the number of emails arriving in your inbox per hour, the number of cars passing a specific point on a highway in a minute, or the number of typos on a printed page.

The Poisson distribution is characterized by a single parameter, $\lambda$ (lambda), which represents the average rate or expected number of events occurring in the interval. This $\lambda$ must be positive ( $\lambda > 0$ ).

A discrete random variable $X$ is said to follow a Poisson distribution with parameter $\lambda$ , denoted as $X \sim \text{Poisson}(\lambda)$ , if its probability mass function (PMF) is given by:

$P(X=k) = \frac{\lambda^k e^{-\lambda}}{k!}$

where:

$k$ is the number of occurrences (a non-negative integer: $k = 0, 1, 2, ...$ )
$\lambda$ is the average rate of occurrence (expected number of events)
$e$ is the base of the natural logarithm (approximately 2.71828)
$k!$ is the factorial of $k$

This formula calculates the probability of observing exactly $k$ events in the interval, given that the average rate is $\lambda$ .

Properties of the Poisson Distribution

The Poisson distribution has some distinct properties:

Mean: The expected value (mean) is equal to the rate parameter: $E[X] = \lambda$ . This makes sense intuitively; the average number of events we expect to see is the rate $\lambda$ .
Variance: The variance is also equal to the rate parameter: $\text{Var}(X) = \lambda$ . This is a unique characteristic. It means that as the average number of events increases, the spread or variability of the number of events also increases.
Shape: The shape of the distribution depends on $\lambda$ . For small $\lambda$ , the distribution is highly skewed to the right. As $\lambda$ increases, the distribution becomes more symmetric and approximates a Normal distribution (a consequence related to the Central Limit Theorem, which we'll discuss later).

When is the Poisson Distribution Applicable?

The Poisson distribution is a good model when the following conditions hold reasonably well for the events being counted:

Independence: Events occur independently of each other. The occurrence of one event does not affect the probability of another event occurring.
Constant Rate: The average rate ( $\lambda$ ) at which events occur is constant over the interval.
Non-Simultaneity: Events cannot occur at the exact same instant.
Proportionality: The probability of an event occurring in a very small interval is proportional to the length of the interval.

Visualizing the Poisson Distribution

Let's visualize the PMF for different values of $\lambda$ . Notice how the shape changes as $\lambda$ increases.

Poisson Probability Mass Function for rate parameters $\lambda = 3$ , $\lambda = 7$ , and $\lambda = 15$ . The distribution shifts right and becomes more spread out and symmetric as $\lambda$ increases.

Applications in Machine Learning and Data Analysis

The Poisson distribution is frequently used in:

Modeling Count Data: Directly modeling variables that represent counts, like the number of clicks on an ad, website traffic, or equipment failures.
Queuing Theory: Analyzing waiting lines, such as customers arriving at a service desk or tasks arriving in a compute queue.
Risk Modeling: Estimating the frequency of rare events, like insurance claims or accidents.
Feature Engineering: Sometimes, Poisson-distributed counts can be used as features in machine learning models.
Poisson Regression: A type of generalized linear model used when the response variable is a count.

Understanding the Poisson distribution provides a foundation for modeling count-based phenomena, which appear often in real-world datasets. Its simplicity, defined by just one parameter $\lambda$ , makes it a useful starting point for many count data problems. We will see how to work with this distribution computationally in Python later in this chapter.