Dealing With Small Datasets In ML

In this project, we'll use different techniques to decrease over-fitting in a machine learning model while keeping high accuracies.

You will learn to:

Effectively deal with small datasets

Learn about regularization

Decrease overfitting

Perform data augmentation

Skills

Machine Learning Fundamentals

Deep Neural Networks

Prerequisites

Intermediate knowledge of Python

Basics of machine learning

Technologies

Python

Tensorflow

Project Description

Machine learning models need a lot of data to train and adjust their parameters. In the case of small datasets, because of the lack of data, it becomes harder to get better results. This issue may lead to overfitting.

In this project, we’ll be given a Sequential model with all of the boilerplate code. This model has around 95% training with 75% validation accuracy, which shows that the model is overfitted.

Throughout the project, we’ll apply different techniques to reduce overfitting while retaining high accuracy.

Project Tasks

Getting Started

Task 0: Introduction

Base Model

Task 1: Overview of Predefined Methods

Task 2: Model Results

Reducing Overfitting

Task 3: Dropout Layer 1

Task 4: Dropout Layer 2

Task 5: Early Stopping

Task 6: Regularization

Task 7: Data Augmentation

Congratulations