# Why can we have misclassifications for a perfect model in logistic regression?

I am reading the book: MACHINE LEARNING- A First Course for Engineers and Scientists, by Lindholm et.al.

Chapter 3, page 50.

Consider the logistic regression for classification.

$$\hat{y}(\mathbf{x})= \begin{cases} &1 & g(\mathbf{x})>r \\\\ &-1 & g(\mathbf{x})\le r & \end{cases}$$

The book says "It can be shown that if $$g(\mathbf{x})=p(y=1|\mathbf{x})$$, that is, the model provides a correct description of the real-world class probabilities, then the choice $$𝑟 = 0.5$$ will give the smallest possible number of misclassifications on average."

I am confused here because if the model provides a correct description of the real-world class probabilities, why do we have misclassifications?

• Could you please give more context for the quote - e.g. chapter name, section name and pages in a specific edition - because this is probably due to what else is assumed part of the problem (e.g. the quote refers to classifying new data, or allows for g(x) for different examples but same vector to be differen - I think it is simply the latter, that $x_i = x_j$ but $g(x_i) \neq g(x_j)$) Nov 11, 2023 at 18:52
• I updated the question to give the book details. Nov 11, 2023 at 19:34
• Are there two reasons? Non separability of classes and some error due to boundary estimation from finite data? Nov 11, 2023 at 20:07
• I am thinking about a very simple example to illustrate both. Classify numbers into positive and negative from uniform distributions in (-1,0) and (0,1) -separable with estimated boundary not exactly zero but close- and overlapping distributions, for example (-3/4,1/4) and (-1/4, 3/4). Nov 11, 2023 at 20:10
• I think book means something different. Nov 11, 2023 at 21:07

Since $$g$$ is your logistic regression model, over data samples $$x$$, the output of $$g(x)$$ is a scalar value between $$0$$ and $$1$$ that is usually interpreted as a probability value.

• We have that $$g(x)=\sigma(W x + b)$$ essentially, where $$W,b$$ are the parameters that you learn and $$\sigma$$ is the sigmoid squashing function (the one that keeps everything in $$[0, 1]$$).
• Furthermore, $$g(x)=p(y=1\mid x)$$ meaning that $$g(x)$$ represents the probability that the sample $$x$$ belongs to the positive or, as the book says in page 49, most probable class - as a side note, this is something that you can change and usually define before learning.

Now, let's assume that the two classes $$y=1$$ (the positive or most probable) and $$y=0$$ (the negative or less probable) have real-world probabilities of: $$p(y=1\mid x) = 0.7$$ and $$p(y=0\mid x) = 1 - p(y=1\mid x) = 0.3$$. Let's also be realistic, in the sense that the two class distributions overlap: this implies that you have some misclassification errors due to the fact that is not possible to separate the two classes perfectly. Note: there could be many reasons for this, such as mis-labelling, and similarity of data samples between the two classes.

At this point, we have $$g(x)=p(y=1\mid x)$$ that is the probability of $$x$$ to belong to the positive class ($$y=1$$), implying that there is a $$70\%$$ chance that $$x$$ has label $$1$$. In order to perform a binary classification, you need to get rid of the probabilities by thresholding them, such as: $$\hat{y}(x) = g(x) > t$$.

• If $$g(x)>t$$ you assign label $$1$$ (so $$\hat{y}(x)=1$$), otherwise you assign label $$0$$.
• In case the model $$g$$ predicts the real-world class probabilities correctly (i.e., $$p(y=1)=0.7$$ and $$p(y=0)=0.3$$) - as assumed by the book - you have that if the threshold is chosen to be $$t=0.5$$ (or $$r=0.5$$ as called in the book), happens that $$g(x)>0.5$$ minimizes the number of misclassifications, equivalently to maximize the accuracy of predictions.
• In other words, for a threshold of $$0.5$$ you get maximum accuracy. But in real classification problems you care also about other measures, such as the AUC of ROC/PR curves, FPR and TPR, F1-score, etc. Basically, you can target a metric and tune the threshold $$t$$ to get the rate of error that best suits you application.

Consider a classifier $$g'(x)=0.7$$ that always outputs a probability of $$0.7$$ regardless the data point, $$x$$. Even if the true class probability is $$0.7$$, this model is still erroneous $$30\%$$ of the times when classifying all the $$x'$$ such that $$y(x') = 0$$.
Note your book specifically claims 'the model provides a correct description of the real-world class probabilities', which exactly means the multi-explanatory variables logistic function $$g(\mathbf{x})$$ trained in your logistic binary classifier $$\hat{y}(\mathbf{x})$$, as a special type of S-shaped sigmoid function transforming the linear combination of input features into a probability between 0 and 1 could be possibly correct in the real world. And if this is true, then intuitively $$r=0.5$$ gives the best overall accuracy and the smallest possible number of misclassifications on average.
In summary it's the logistic model $$g(\mathbf{x})$$ could possibly provide a correct description of the real-world class probabilities, but even this is the case, the resulting binary classifier $$\hat{y}(\mathbf{x})$$ would still possibly have misclassification since its accuracy may not be 100% either on your training samples or unseen test samples.