Cursusinhoud
Recognizing Handwritten Digits
Class Balance
Ensuring class balance in machine learning datasets is crucial for avoiding model bias towards any specific class. When a dataset is balanced, it signifies an equitable representation of all classes, which is vital. An imbalanced dataset can result in suboptimal performance of the model, especially in predicting the minority class.
Consider a dataset comprising customer transactions where 90% are legitimate and only 10% fraudulent. Training a model on such skewed data might incline it to predict most transactions as legitimate. This inclination occurs because the model is tailored to reduce overall error, and identifying most transactions as legitimate boosts accuracy, albeit superficially.
Hence, maintaining class balance is imperative for training models on a diverse sample from each class, enhancing their ability to make precise predictions across the board.
Swipe to start coding
-
Extract the target labels (digit values) from the MNIST dataset and assign them to variable
Y. -
Visualize the frequency distribution of digit labels from the MNIST dataset by creating a count plot.
Oplossing
Schakel over naar desktop voor praktijkervaringGa verder vanaf waar je bent met een van de onderstaande optiesBedankt voor je feedback!
