Strategies to Improve the Model

Introduction

Till now, we have created various projects using the Custom Vision service. For the sake of simplicity, we have used a very small dataset to train our models. But, in real-life scenarios, we would need to use very large datasets and many times the model may be able to perform well.

In this lesson, we’ll discuss some of the best practices that can be used to improve our model’s performance.

Avoiding overfitting of the model

Overfitting occurs when the model has been trained with a limited variety of data and starts giving importance to features that are not required. For example, if we’re building an image classifier which will classify the images into “apple” or “citrus” as shown in the below image.

Here, the model may learn that anything that contains “hand” could be an “apple” and anything that contains “plates” could be a “citrus”.