Download presentation
Presentation is loading. Please wait.
Published byHarry Baumhauer Modified over 5 years ago
1
Course Lab Introduction to IBM Watson Analytics
University of Rome «La Sapienza» Course of Business Intelligence Course Lab Introduction to IBM Watson Analytics
2
Ing. Ivonne E. Vereau Tolino
Lab Speakers Ing. Vittorio Carullo Software Architect IBM Watson Squad Senior Member Ing. Ivonne E. Vereau Tolino Software Engineer IBM Software Services Specialist
3
Target & Scope Familiarize with a «real» software used in large enterprises Accomplish small but significant use cases in BI arena Understand the impact of tools over team productivity Introduce advanced topics like the use of “non structured” information Lab sessions will be held on Thursday, starting from September 27, 2018 , pm
4
Labs Schedule Presentation of the tool and its basic features
Lab 1: Introduction to Watson Analytics Understanding and use of the tool features or conducting BI use cases Lab 2: Working with Data Lab 3: Analyze and Discover Lab 4: Predict and take decisions Lab 5: Report and visualization Use of the tool for Advanced Analytics Lab 6: Working with Social Media Lab 7: Introduction to Content Analytics Lab 8: Putting all together Today’s topic is highlighted!
5
Reference Materials Explore IBM Watson Analytics library/analytics/watsonanalyticsgallery/ A gallery of use cases with related datasets and supporting data visualizations IBM Knowledge Center utions.wa.doc/welcome.html A technical reference for product features
6
Today’s Contents: Analyze and Discover (Part 2)
Predictive analysis Decision rules and Decision tree
7
1. Predictive Analysis
8
Identify key drivers When you select a Starting Point or submit a question within the Discovery set, Watson analytics will build a linear regression model to quantify the impact that each field and each potential combination of fields, has on our dependent variable Dependent variable = target Strength essentially captures how well each driver can explain the variance in our target Predictive power! Drivers = Factors impacting the target
9
The “spiral” visualization
After creating the Discovery, you will be able to navigate the data from different perspectives using a visualization. To best identify the predictive aspects of the data, it is recommended to use the Spiral Visualization The Spiral Visualization opens displaying a target in the center of the spiral, with icons surrounding it.
10
The “spiral” visualization (cont.)
You can identify the factors impacting the key performance indicator, by hovering over the surrounding icons in the spiral. In some cases, there is a combination of factors having impact, or there might be only a single factor. Combination factors and single factors are represented by different icons. Factors closes to the center of spiral have the highest predictive strength than a single factor.
11
The “spiral” visualization (cont.)
12
Target to analyze You can change target
When changing target, also Drivers change.
13
Analysis of factors Now, that you know which factors are predictors of the target, you can continue your analysis of the factors using the table.
14
Predictive analysis: Hands on
Upload spreadsheet into Watson Analytics «WA_Fn UseC_ Marketing Campaign Eff UseC_ FastF.csv» Let’s look at the data asset and understand the meaning of columns
15
Predictive analysis: Hands on
Pose a natural question «What drives SalesInThousands?» Choose Spiral visualization Within Discovery Set, identify Drivers Change the target from «SalesInThousands» to «MarketSize» Identify new Drivers
16
2. Decision rules and Decision tree
17
How to view decision rules and the decision tree in a predictive model?
To analyse the predictive factors of a specific target choose a Data Asset. Based on the dataset, Watson Analytics generates a predictive model that includes: Associated Decision Rules Decision Tree visualizations
18
Decision rules Decision rules are a set of statistically generated profiles, with each profile showing you a group of factors that are used that to classify records.
19
Decision rules (cont.) This classification helps identify which combination of factors are probable to result in a specific outcome for the target field
20
Decision rules (cont.) The Decision rules tab also provides an overall predictive strength for all the records in the data set.
21
Decision rules (cont.) In this case, the first profile predicts an outcome with the highest average case call duration. This outcome is predicted to occur when Agent Training Level is No Training and Case Type is Request and when Case Area is Hardware.
22
Decision rules (cont.) You can also identify what combination of factors will predict an outcome with the lowest average case call duration. This can be done by changing the sort on the predicted value column, from the current descending, to ascending.
23
Decision tree The decision tree shows you patterns of characteristics that lead to a certain outcome, which you can think of as profiles.
24
Decision tree (cont.) Reading from left to right, each branch in the tree is a unique pattern that leads to the likelihood of an outcome occurring in the past. Each level of the tree (from left to right) has a higher predictive strength than the subsequent levels and branches.
25
Decision tree (cont.) You can examine a pattern by collapsing nodes to their lowest leaf level, and then expanding the nodes to analyze predicted outcomes at each level.
26
Decision tree (cont.) The tree also includes a color gradient at each level to emphasize the target values. As shown by the gradient legend, a darker color indicates higher average values, while a lighter color indicates lower values.
27
Decision tree (cont.) At the end of the branch, you can identify that the lowest level in this branch is Agent Training Level, and the training level with the highest average case call duration is No training.
28
Decision tree (cont.) As you look back through the branch, you can identify that the levels and values showing the highest average call duration, are the same factors, profile, and values you previously identified on the Decision rules tab.
29
Predictive analysis It is up to you as to which view you use to interpret decision rules. Typically, most users prefer the natural language view of the Decision rules tab. And what about you, what do you prefer?
30
Predictive Analysis: Hands on
Upload “Service Agent Performance dataset” to Watson Analytics Let’s look at the data asset and understand the meaning of columns
31
Predictive Analysis: Hands on
Within the Decision rules tab, analyse the target “Case Call Duration”, check the profiles and tell what is the factor that drives the greatest duration call. Within the Decision tree, using the color gradient, identify which case type has the highest average case call duration. Change target to analyse, for example: Service Satisfaction and check what it changes.
Similar presentations
© 2025 SlidePlayer.com. Inc.
All rights reserved.