Book Image

Data Forecasting and Segmentation Using Microsoft Excel

By : Fernando Roque
Book Image

Data Forecasting and Segmentation Using Microsoft Excel

By: Fernando Roque

Overview of this book

Data Forecasting and Segmentation Using Microsoft Excel guides you through basic statistics to test whether your data can be used to perform regression predictions and time series forecasts. The exercises covered in this book use real-life data from Kaggle, such as demand for seasonal air tickets and credit card fraud detection. You’ll learn how to apply the grouping K-means algorithm, which helps you find segments of your data that are impossible to see with other analyses, such as business intelligence (BI) and pivot analysis. By analyzing groups returned by K-means, you’ll be able to detect outliers that could indicate possible fraud or a bad function in network packets. By the end of this Microsoft Excel book, you’ll be able to use the classification algorithm to group data with different variables. You’ll also be able to train linear and time series models to perform predictions and forecasts based on past data.
Table of Contents (19 chapters)
1
Part 1 – An Introduction to Machine Learning Functions
5
Part 2 – Grouping Data to Find Segments and Outliers
10
Part 3 – Simple and Multiple Linear Regression Analysis
14
Part 4 – Predicting Values with Time Series

Pivot analysis of the outliers

We can apply the business intelligence pivot tables to explore the ranges of the groups for every variable of the dataset. Using this method, we can visualize the groups that appear to be outliers.

Kaggle credit card fraud dataset

With the information of the group assignment with K-means clustering, we can explore the outliers for each dataset. From the amount chart of credit card transactions in Figure 7.9, we see that groups three and four have compact and similar values with a combined range between 355 and 1402:

Figure 7.9 – Credit card amount field groups

From Figure 7.9, we could conclude that the possible outliers are as follows:

  • Group 1 (ranges between 0 and 86)
  • Group 5 (ranges between 89 and 322)
  • Group 4 (has just one record with a big value of 3828, which indicates an anomaly)

Combining the analysis with the V1 field groups, in Figure 7.10, we can examine whether we can confirm the...