ThingWorx Analytics Training: Module 5 Part 1

Chapter Summaries

0:01 – Introduction: Module 5 Descriptive Analytics in ThingWorx Analytics This opening frames Module Five and previews how to perform descriptive analytics in ThingWorx Analytics using signals, profiles, and clustering to derive relations and patterns from your data.

0:28 – ThingWorx Analytics Data Flow: Feature Engineering to Champion Model Walks through the analytics-ready pipeline — raw source data, data profiling, feature engineering, and metadata — then loading into ThingWorx to iterate models and select a best-performing champion model for smart, connected applications.

1:43 – Analytics Continuum: Descriptive, Predictive, Prescriptive, Persuasive & Automated Explains the five stages of analytics maturity, from descriptive analytics on historical data and KPIs through predictive, prescriptive, persuasive, and fully automated decision analytics that act with little human intervention.

4:38 – Continuum Examples: Predictive Maintenance, Fracking & Retail Upsell Grounds each continuum stage in real use cases: dashboards for descriptive insight, loan-default and machine-maintenance predictions, prescriptive drilling site selection in oil and gas, and persuasive retail upsell like suggesting maple syrup with pancake mix.

5:55 – Signals & Correlation: Pearson Correlation Coefficient Introduces Signals, the first descriptive service, and covers correlation between continuous variables using the Pearson correlation coefficient, which ranges from negative one to one to measure linear relationship strength.

7:12 – Reading Correlation Graphs: +1, -1, 0.3 and Zero Coefficients Uses scatter-plot examples to interpret positive, negative, weak (0.3), and zero correlation coefficients, showing how tighter point bands indicate stronger linear relationships between X and Y variables.

8:26 – Why Correlation Matters: Feature Selection for Predictive Accuracy Explains why correlation drives feature selection — training on features with strong correlation to the goal variable narrows the range of possible Y values and improves a machine learning model's predictive accuracy.

9:49 – Non-Linear Relationships & Pearson Limitations (Categorical Variables) Warns that a strong relationship isn't always linear, using book thickness versus the temperature-and-ice-cream-sales curve, and flags Pearson's limits with non-linear patterns and non-numeric data like strings and categoricals such as petal color.

12:01 – Mutual Information: Measuring Predictive Relationships Defines mutual information, the metric behind ThingWorx Analytics Signals that quantifies how much one variable reveals about another, capturing predictive relationships across data types beyond numeric values, illustrated with an age-and-income Venn diagram.

12:49 – ThingWorx Signals from Mutual Information & Feature Reduction Details how Signals rank features by normalized mutual information (0 to 1) against the goal variable, sorted highest to lowest, enabling feature reduction and informing which features are included in ThingWorx Analytics models.

13:52 – Multicollinearity & the Redundancy Filter Explains multicollinearity, where highly correlated independent variables (like years in college versus highest degree) become redundant, and introduces the Redundancy Filter checkbox in Analytics Builder and ThingWorx Services to filter redundant features.

15:47 – Demo: Building a Signals Job in Analytics Builder (BeanPro Dataset) Step-by-step demo of creating a Signals job in ThingWorx Analytics Builder with the BeanPro dataset — naming the job, enabling the Redundancy Filter, setting the grinder error occurrence goal, using Exclude Features and filters, then submitting.

17:13 – Signals Job Results: MI Bins, Difference vs. Average & CSV Export Reviews Signals results on BeanPro data — feature sorting with and without redundancy filtering, value bins, record counts, average goal rates, difference and percentage-difference versus the 18.66% overall failure rate, the outlier slider, search, and CSV export.