Logistic Regression and the Classification Task of Machine Learning [Lecture 2.1]

"How to derive the logistic regression from linear regression and use it for classification. Why not use linear regression model for classification?"

___________________________________________

___________________________________________
Part 1: Logistic Regression and the Classification Task of Machine Learning

Part 2: Classification with the k-Nearest Neighbor Algorithm - kNN

Part 3: Softmax Regression as a Generalization of Logistic Regression for Classification

Part 4: One vs One and One vs All for Multiple Class Classifications of Machine Learning Models

Part 5: Cross Entropy vs. MSE as Cost Function for Logistic Regression for Classification

Part 6: What is the Meaning of Cross Entropy/ Log Loss as Cost Function for Classification

Part 7: How to Evaluate Classification Models - Confusion Matrix and Precision-Recall Curve

___________________________________________
The classification task of machine learning is introduced. Examples of handwritten digit recognition and spam mail detection are considered. The Locistic regression model is presented to tackle the classification task. This model bases on the logistic function which is also called sigmoid function. The output of the sigmoid function gives the probability for the classes. The logistic regression model needs to specify a decision boundary. Further, the model is trained by a cost function: the log loss/cross entropy. This cost function is used to fit the model to data.
Finally, the question why not unsing linear regression is answered: functions with limited value rage are preffered for classification.