From the course: Mistakes to Avoid in Machine Learning
Unlock the full course today
Join today to access over 22,600 courses taught by industry experts or purchase this course individually.
Overfitting your models
From the course: Mistakes to Avoid in Machine Learning
Overfitting your models
- [Instructor] Arguably the most common technical mistake in machine learning is referred to as overfitting. Overfitting is when your model captures patterns in your training data too well. Meaning it doesn't generalize well to unseen data. Essentially your model has become highly attuned to the noise in your training set rather than the signal. So here's an example, let's say you wanted to predict which of your coworkers would go out to lunch versus bring one from home. Eventually, after memorizing the car of each of your coworkers, you learned who went to lunch by looking out into the parking lot. Thus achieving a model with great predictive accuracy on your training set. However, this model would be highly confused if you introduced a new coworker. And your prediction would be meaningless. So if you see a large drop-off in performance between your train and test groups, your model is likely overfit. So how do…
Practice while you learn with exercise files
Download the files the instructor uses to teach the course. Follow along and learn by watching, listening and practicing.
Contents
-
-
-
Assuming data is good to go2m 2s
-
(Locked)
Neglecting to consult subject matter experts1m 48s
-
(Locked)
Overfitting your models3m 25s
-
(Locked)
Not standardizing your data2m 57s
-
(Locked)
Focusing on the wrong factors2m 11s
-
(Locked)
Data leakage2m 40s
-
(Locked)
Forgetting traditional statistics tools1m 57s
-
(Locked)
Assuming deployment is a breeze1m 47s
-
(Locked)
Assuming machine learning is the answer1m 35s
-
(Locked)
Developing in a silo2m 16s
-
(Locked)
Not treating for imbalanced sampling3m 29s
-
(Locked)
Interpreting your coefficients without properly treating for multicollinearity3m 19s
-
(Locked)
Evaluating by accuracy alone6m 8s
-
(Locked)
Giving overly technical presentations1m 56s
-
-