AIToday
Qiita 機械学習Published: Nov 18, 2026, 13:00 JST

分類問題 explained: thresholds, imbalance and G検定 prep

Send on LINE
分類問題 explained: thresholds, imbalance and G検定 prep

3 Key Points

  1. What happened

    The explainer defines 分類問題 as assigning inputs to predefined categories, split into binary tasks (spam detection, disease diagnosis) and multiclass tasks (image classification), using ロジスティック回帰, SVM, ナイーブベイズ, decision trees and neural networks.

  2. Why it matters

    For binary classification it warns the default threshold of 0.5 is not always optimal, and recommends searching for a better cutoff with PR curves or ROC curves, while tackling class imbalance through class weighting, resampling, or SMOTE.

Not sure about something? Ask the AI

Questions and answers are published on this page.

Summaries like this, every morning.

Context & Analysis

The piece is structured as a study guide rather than a news report, and its framing reflects that: it moves from definitions to model families to implementation cautions, then closes with exam preparation. The treatment of thresholds is notable because it treats the 0.5 cutoff as a starting point rather than a rule — a point that matters for binary tasks like spam detection and disease diagnosis, where the cost of the two error types differs. On imbalance, class weighting, oversampling and undersampling, and SMOTE are grouped as options, and implementation can combine class weights, cost-function adjustments and sampling methods.

The inclusion of probability calibration sits alongside these cautions, with CalibrationCurve named as the evaluation tool, suggesting the guide treats predicted probabilities as something to verify rather than trust by default. The G検定 section is presented as a practical tie-in: the topic is described as frequently appearing on the exam and listed in the syllabus, and the closing resources lean on official and commercial textbooks plus a free practice-question site. Taken together, this is best read as a beginner's orientation to classification — one that flags the places where defaults can mislead rather than presenting the task as settled once a model is chosen.

FAQ
What is the difference between binary and multiclass classification?
Binary classification assigns inputs to one of two categories, such as spam detection or disease diagnosis, and typically uses logistic regression. Multiclass classification handles three or more categories, such as images of dogs, cats and birds, and uses the softmax function.
Why is the default 0.5 threshold not always best?
The default threshold of 0.5 is not necessarily optimal, especially in binary classification. It recommends using PR curves and ROC curves to search for the best threshold.
Class imbalance can be handled with class weighting, oversampling and undersampling, and SMOTE.
It lists class weighting, oversampling and undersampling, and SMOTE. In implementation, it suggests combining class weights, cost-function adjustments and sampling methods.
Qiita 機械学習Read Original Article
Send on LINE

AI news that matters for your work, delivered every morning.

Pick your industry and the AI tools you use, and get news related to your work every day.

Free · 30 seconds with Google · unsubscribe anytimeWhat is AIToday? →

Ask AI

Ask AI anything about this article. The AI reads this article, earlier AIToday articles, and Wikipedia, and cites its sources. Q&As are published on this page for other readers too.

Questions and answers are published on this page.

Next articleTricentis: 65.6% in Japan ship untested code as quality losses top $1 million