I just completed Module 3: Classification from ML Zoomcamp 2026 🎉

This module was about getting into ML with Classification

Link: colab.research.google.com/drive/1QBQ8i...

#mlzoomcamp @Alexey Grigorev

Today I tested different regularization settings for Logistic Regression and compared their validation accuracy. Small parameter changes can lead to different model performance.
#mlzoomcamp #DataScience #LearningInPublic

I experimented with removing features from a classification model and comparing validation accuracy. Sometimes removing a feature barely changes the result.
#mlzoomcamp #MachineLearning

Today’s practice: categorical data → one-hot encoding → Logistic Regression → predictions → accuracy. Nice to see the whole classification workflow coming together.
#mlzoomcamp #Python

Learned about Mutual Information today. It gives us another way to understand how useful a categorical variable may be for predicting a target.
#mlzoomcamp #LearningInPublic

Today I practiced creating train, validation, and test datasets. Training teaches the model, validation helps us evaluate choices, and testing is saved for the final evaluation.
#mlzoomcamp #MachineLearning

One lesson from today: correlation helps us measure the linear relationship between numerical features. I practiced comparing feature pairs using a correlation matrix.
#mlzoomcamp #DataScience

Learning more about classification today. I practiced checking a dataset for missing values and preparing categorical and numerical columns for machine learning.
#mlzoomcamp #MachineLearning

Finished another #mlzoomcamp homework! 🎉 This time I practiced correlation, mutual information, one-hot encoding, logistic regression, accuracy, feature elimination, and regularization.
#MachineLearning #LearningInPublic

Today I experimented with feature elimination: remove one feature, retrain the model, and compare the accuracy. A simple way to see which features matter less.
#mlzoomcamp #DataScience

Practiced one-hot encoding categorical variables and training a Logistic Regression model today. Turning categories into features finally makes more sense!
#mlzoomcamp #Python #MachineLearning

Today I learned about Mutual Information and used it to see which categorical feature contains the most information about the target variable.
#mlzoomcamp #LearningInPublic

Practiced splitting data into training, validation, and test sets today. I’m starting to understand why keeping the test set separate is so important.
#mlzoomcamp #MachineLearning

Today I learned how to create and read a correlation matrix. It was interesting to see how strongly some numerical features are related to each other.
#mlzoomcamp #DataScience

Today I practiced data preparation for classification. I checked missing values and learned how to handle missing categorical and numerical features before training a model.
#mlzoomcamp #MachineLearning

7/ Rather than settling for a hard 0 or 1 prediction, using soft predictions outputs the actual probability. This allows businesses to apply their own threshold to decide when to intervene with a promotional email to retain the customer. #mlzoomcamp

6/ Logistic regression is highly interpretable. Each one-hot encoded category gets its own weight: a positive weight pushes the predicted churn probability up, while a negative weight pushes it down. #mlzoomcamp

5/ Machine learning models only work with numbers, making categorical string values useless on their own. One-hot encoding solves this by replacing a categorical column with several binary columns, turning each category into a 0 or 1. #mlzoomcamp

4/ For numerical features, the Pearson correlation coefficient measures the degree of dependency with the target. A negative correlation, like with customer tenure, means the longer a customer stays, the less likely they are to churn. #mlzoomcamp

3/ Measuring the importance of categorical variables requires mutual information. This concept from information theory tells us exactly how much we learn about the target churn variable by knowing a specific feature's value. #mlzoomcamp