ThingWorx Analytics Training: Module 5 Part 2

Chapter Summaries

0:00 – Profiles: Introduction & Definition in ThingWorx Analytics Introduces profiles as a descriptive analytics method in ThingWorx Analytics, defining a profile as a statistically distinct subpopulation and explaining how it identifies data hotspots for groups that over- or underperform relative to the goal variable.

0:38 – Profiles Business Value: Machine Failure Hotspot Example Illustrates profiles with a machine-failure scenario, showing how location, time of day, and a specific machine model combine to drive failures, delivering interpretable, business-actionable insight without needing a predictive model.

1:09 – Beam Search & Decision Trees: Z-Scores & Beam Width Explains how profiles use a beam search to build a decision tree across Boolean and categorical features, computing a Z-score of statistical significance for each branch and using beam width to retain the highest-scoring paths.

2:39 – Profile Job Parameters: Max Features, Min Population % & Mutual Exclusivity Covers when the search stops — at max depth or the minimum population percentage — how each beam search yields one profile ranked by Z-score, and why profiles are mutually exclusive so each feature value lands in only one profile.

3:41 – Binning Continuous Values in Profiles Jobs Describes automatic binning of continuous values, noting the bin count is system-determined and non-configurable, and how adjacent bins with similar high Z-scores are merged together to make the results more meaningful.

4:01 – Creating Profiles in Analytics Builder: Job Setup & Configuration Demonstrates building a profiles job in ThingWorx Analytics Builder — naming the job, selecting or loading data, Exclude Features, maximize/minimize goal, Z-score vs distance-from-mean calculation, Minimum Population Percent, Number of Profiles, and Max Depth.

5:44 – Interpreting Profiles Results: Grinder Failure Rates & Export Walks through reading profile results using grinder-failure data — average daily cycles, satellite grinder type, average daily pressure — showing failure rates, difference-versus-average percentages, record counts, and exporting results to tabular format.

8:29 – Clustering Introduction: Supervised vs Unsupervised Learning Introduces clustering as the next descriptive analytics topic, contrasting supervised learning (signals and profiles using a labeled goal variable) with clustering's unsupervised approach that finds patterns in unlabeled data on its own.

9:24 – How Clustering Works: Intra/Inter-Cluster Distance & Outliers Explains clustering's aim to minimize intra-cluster distance and maximize inter-cluster distance to uncover natural groupings, and why the mean-based cluster centers used in ThingWorx Analytics are sensitive to outliers.

10:52 – Hierarchical vs Partitional Clustering & K-Means Compares hierarchical clustering (dendrogram trees, no preset cluster count, ideal for categorical data) with partitional clustering, focusing on the K-means algorithm that assigns observations to K clusters by their nearest mean.

13:29 – Profiles vs Clustering: Explainability & Goal Variable Compares profiles and clustering — profiles are more explainable, focus on high-impact hotspots, and use the goal variable, while clustering covers every data point through unsupervised learning and reveals overall similarity.

14:48 – Creating Signals & Profiles Jobs: Services, Mashups & Builder Reviews the ways to create signals and profiles jobs — via ThingWorx Services called in code or a ThingWorx mashup, or in Analytics Builder (standalone or within a model) — all running identical algorithms, then closes the module.