Book Image

Codeless Time Series Analysis with KNIME

By : KNIME AG, Corey Weisinger, Maarit Widmann, Daniele Tonini
Book Image

Codeless Time Series Analysis with KNIME

By: KNIME AG, Corey Weisinger, Maarit Widmann, Daniele Tonini

Overview of this book

This book will take you on a practical journey, teaching you how to implement solutions for many use cases involving time series analysis techniques. This learning journey is organized in a crescendo of difficulty, starting from the easiest yet effective techniques applied to weather forecasting, then introducing ARIMA and its variations, moving on to machine learning for audio signal classification, training deep learning architectures to predict glucose levels and electrical energy demand, and ending with an approach to anomaly detection in IoT. There’s no time series analysis book without a solution for stock price predictions and you’ll find this use case at the end of the book, together with a few more demand prediction use cases that rely on the integration of KNIME Analytics Platform and other external tools. By the end of this time series book, you’ll have learned about popular time series analysis techniques and algorithms, KNIME Analytics Platform, its time series extension, and how to apply both to common use cases.
Table of Contents (20 chapters)
1
Part 1: Time Series Basics and KNIME Analytics Platform
7
Part 2: Building and Deploying a Forecasting Model
14
Part 3: Forecasting on Mixed Platforms

Chapter 9: Training and Deploying a Neural Network to Predict Glucose Levels

In this chapter, we will look at a more critical prediction problem: forecasting glucose levels to provide diabetics with early warnings when their insulin or carbohydrates need to be balanced due to their blood sugar levels.

We will also introduce neural networks. We will start by learning how to use a neural network for time series prediction by using the simplest neural architecture: a classic feedforward neural network (FFNN) without any recurrent layers. Though simple, the results obtained by this neural network for glucose level prediction are already quite accurate. However, the network’s performance could be improved by using a more complex network architecture, such as long short-term memory (LSTM), which we will introduce in the next chapter.

Thus, we will use this glucose level prediction case study to briefly explain how neural networks work and how they can be trained. Finally, we...