ThingWorx Analytics Training: Module 7 Part 1
This video begins Module 7: Predictive & Prescriptive Scoring of the ThingWorx Analytics Training videos. It describes how a trained machine learning model takes inputs and makes predictions of different kinds, depending on the use case. It shows how scoring works in production, taking inputs from various sources and producing a score to help users make informed decisions. It also covers the concept of field importance in an individual score.
This video contains a hands-on exercise, and it is recommended to complete it before moving on to subsequent sections. In this exercise you will use the model you trained in Module 6 to perform a batch scoring job.
This video has chapters — jump to any section using the chapter menu in the video player.
Chapter Summaries
0:00 – Module 7 Overview: ThingWorx Analytics Data-to-Insight Pipeline
Introduces Module Seven on predictive and prescriptive scoring, explaining how ThingWorx Analytics ingests raw source data from social, client, text file, spreadsheet, or database sources into a contextual repository structured as a dimensional model with facts and dimensions.
0:54 – Feature Engineering & Analytics Inputs: Signals, Clusters & Profiles
Covers the manual feature engineering of static and time-series features performed outside ThingWorx Analytics, then loading the structured data to run descriptive and predictive analytics that surface relations and patterns as signals, clusters, and profiles.
1:37 – Prescriptive Action Plans: Predictive Maintenance & Part Optimization Use Cases
Shows how profiled audiences and tactical action plans turn predictions into action, answering business questions like when to service parts and how to make lower-performing machines behave more like higher-performing ones.
2:16 – ThingWorx Analytics Data Flow: Explore Data, Scope Use Case & Structure
Walks through the data flow journey — exploring raw data for completeness and coverage, scoping a use case around assets or production lines, and structuring the data to support feature engineering and measurable business value.
3:08 – Metadata Detection, Analytics-Ready View & Iterative Champion Model
Explains automatic metadata detection, constructing the analytics-ready view, feeding it into the ThingWorx analytics engine, and the iterative loop of building and validating predictive models until a champion model is deployed to a smart connected mashup.
4:05 – The Analytics Continuum: Descriptive Business Intelligence to Predictive Analytics & Machine Learning
Presents the analytics maturity curve, contrasting descriptive analytics or business intelligence (answering what happened and why) with predictive analytics powered by machine learning that answers what will happen next.
5:08 – Prescriptive, Persuasive & Automated Analytics with an IoT Smart-Home Example
Extends the continuum through prescriptive analytics simulations, persuasive analytics like streaming and product recommendations, and automated analytics illustrated with a connected fitness tracker, smart thermostat, and smart security system delegating decisions.
6:48 – Predictive Scoring Explained: Coefficients, Validation Sets, Boolean Risk & Thresholds
Defines predictive scoring as an equation with estimated coefficients or weights, scoring validation-set records, and returning Boolean goals as a risk value between zero and one compared to a threshold ThingWorx Analytics auto-tunes for best model accuracy.
8:13 – Scoring in Production: Batch vs Real-Time Scoring & the Mashup Feedback Loop
Details how a published champion model scores data collected from multiple machines, comparing batch scoring for thousands of records against event- or scheduler-triggered real-time scoring, then feeding predictions to mashups and back into ThingWorx for model updates.
9:56 – Coffee Machine Failure Use Case: Regression Equation & Denormalized Predictions
Explores predicting the number of days until a coffee-machine part fails using a regression equation with independent variables, beta coefficients, an intercept term, and an error term (epsilon), denormalizing the continuous output into a numeric days-to-failure value.
10:57 – Field Importance: Per-Record Predictive Influence & Industry Applications
Introduces field importance as a per-record ranking of a field's predictive impact or influence, illustrated through fitness trainers, medical risk factors, marketing prospects, and production-line technicians proactively addressing high machine-failure risk.
12:33 – Calculating Field Weights: Jack's Cancer Risk & Cigarette Variation Example
Explains extracting a field's relative weight from a trained model using Jack's cancer-risk scenario, varying cigarettes smoked per day to move risk between thirty and fifty-five percent, and distinguishing high field importance from low field importance.
14:56 – Measuring Feature Weight: Full Range, Distance from Max & Distance from Min
Compares the three ThingWorx Analytics methods for measuring feature weight — full range (maximum minus minimum risk), distance from max, and distance from min — for understanding total impact and whether to increase or decrease a prediction.
16:26 – Batch Scoring Demo: BeanPro Model in ThingWorx Analytics Builder
Demonstrates a hands-on batch scoring job in Builder using the BeanPro model — creating a scoring data filter on Record Purpose, selecting the Distance from Min causal technique, adding an Identifier field, submitting the job, and interpreting the downloaded score and feature-weight columns.