Методы и оценка машинного обучения

Уровень 0 · материалов: 4

В кластер входят документы, посвященные применению, настройке и оценке прогностических моделей машинного обучения, и не входят документы об алгоритмах поиска сообществ в графах или методах визуализации данных.

Общие признаки: алгоритмы машинного обучения, прогнозирование поведения клиентов, метрики оценки моделей, оптимизация моделей

Группа выше: Методы обучения и оптимизации моделей

Материалы

• #3225 — Introduction to Uplift Modeling for Marketing Optimization — Представитель этого кластера

  Туториал по uplift моделированию. Часть 1 — maks-sh

  Смысл: The text introduces the concept of uplift modeling in marketing, explaining how to isolate the true effect of a communication by comparing a treatment group to a control group. It categorizes customers based on their reaction to stimuli and presents various algorithmic frameworks to predict which users are most likely to be 'persuaded' by an offer.

  A technical guide on using uplift modeling to identify 'persuadable' customers by calculating the causal effect of marketing communications through treatment and control group analysis.

• #24086 — Building a Neural Network for Lead Scoring Tasks

  Краткий курс машинного обучения или как создать нейронную сеть для решения скоринг задачи — Dirac

  Смысл: The main idea is that machine learning, specifically neural networks, can replace manual, hypothesis-based business scoring models to more accurately predict customer behavior and optimize sales efforts through data-driven weight optimization.

  A practical guide on implementing a neural network for lead scoring, covering everything from data preprocessing and network theory to model evaluation and business integration.

• #17828 — Metrics in Machine Learning Classification Tasks

  Метрики в задачах машинного обучения — egor_labintcev

  Смысл: The main idea is that selecting the correct evaluation metric is crucial for machine learning success, as a single metric like Accuracy can be deceptive; practitioners must choose metrics (like Precision, Recall, AUC-PR, or Logloss) based on the class balance and the specific costs of False Positives versus False Negatives in their business context.

  A technical guide on choosing and implementing classification metrics in machine learning, emphasizing why accuracy fails for imbalanced data and how to use precision, recall, and AUC curves instead.

• #29810 — Optimizing Random Forest with PCA and Hyperparameter Tuning

  Random Forest, метод главных компонент и оптимизация гиперпараметров: пример решения задачи классификации на Python — ru_vds

  Смысл: The main idea is to demonstrate that adding complexity to a machine learning model—such as dimensionality reduction via PCA and exhaustive hyperparameter tuning—does not always lead to better performance, as shown by the baseline Random Forest model outperforming the tuned versions in a breast cancer classification task.

  The article compares a baseline Random Forest model against versions using PCA and hyperparameter tuning, finding that the simpler baseline model achieved the highest recall for breast cancer detection.