In “Is a Tesla in Your Future?” Part 1 and Part 2, we used Oracle Machine Learning to build ML models to predict which customers were most likely to purchase Tesla cars based on historical data we have collected on their demographics and past Tesla buyers.  However, this is only some of the data we could use to predict Tesla buyers.   What if we knew that someone lives next to Elon Musk or two of their four neighbors own Tesla cars?   What if we could add review comments about their experiences renting or test-driving Tesla cars?   Shouldn’t their location data be a good predictor?   Perhaps they live or spend time in locations where battery life or supply of charging stations may be a concern.   All these additional data sources and customer points of view can help to provide better, more insightful predictive ML models.

See “Is a Tesla in Your Future? A Machine Learning Approach using Structured, Unstructured, Relationship and Location Data” presentation.

Download the on-premises version presentation that uses Oracle Data Miner (click to download ODMr workflow) and Oracle Analytics Cloud.

In Part 3, we take advantage of the wide range of features that are supported in Oracle Databases and add 3 different data types to create a more complete 360-degree customer point of view (POV):

– Tesla reviews (unstructured data)   


– Tesla friends (relationship / “graph” data)


– Telsa location data (geospatial / transactional location data)


The Oracle “Converged” Database is a multi-model, multi-tenant, multi-workload architecture that delivers the data models and access methods needed for modern development on all workloads (OLAP, OLTP, IoT, etc.). Better, it delivers the specific data (i.e. structured, unstructured, transactional) type support, analytical, machine learning, graph and location functions, security, integration, and operational characteristics developers, data analysts, data scientists, and developer teams are seeking to collaborate together — powerful, smart capabilities that simplify, streamline, and out-perform at scale.

Oracle Autonomous Data Warehouse – Integrated Cloud Data Service to Empower Business Innovators

We continue forward using our Oracle Machine Learning notebook, make a duplicate notebook and start adding paragraphs to view and later join our three additional data sources:  

– Tesla reviews (unstructured data)   

– Tesla friends (relationship / “graph” data)

– Telsa location data (geospatial / transactional location data)

Our Tesla reviews data is unstructured i.e. text and must be defined as CLOB data type for Oracle Machine Learning. See Machine Learning Operations on Unstructured Text.  

We also need to define our lexer, tokenization, and stopwords for Oracle Text to preprocess the unstructured data for Oracle Machine Learning.

Because some of our data tables are transactional (location data and friends data), we need to first  aggregate the transactions and then join them with the Telsa78 customer demographics and Tesla reviews (unstructured data.)

Oracle Machine Learning identifies the most influential input attributes including demographic data, reviews (unstructured data), friends (“graph” relationship data) and location data (geospatial data) that most predict Tesla buyers.  

We continue forward by building multiple Oracle Machine Learning models using our enhanced 360-degree customer view.

We compare model performance using the test data.

We join the model evaluations together and can compare many model metrics including accuracy, lift, AUC and ROC.

Now we can use our “best” model and apply it (100% inside ADW) to make predictions about which customers are most likely to buy a new Tesla car.

Oracle Machine Learning provides several options for model deployment including:

Lastly, we create a LIKELY_TESLA_BUYERS table in the database for access to other applications and users.   

Next, we analyze our insights and predictions about likely Tesla buyers using Oracle Analytics Cloud in Is a Tesla in Your Future? – Part 4.

Download the fun Tesla Demo Artifacts (Tesla demo datasets TESLA_78, TESLA_REVIEWS89, TESLA_FRIENDS89, TESLA_LOCATIONS89, Is a Tesla in Your Future_ A 360-degree POV using OML OML notebook, and Tesla Buyers OAC Project.dva)

– Naman Mehta Principal Member of Technical Staff who worked diligently to get all the complex SQL joins, text mining components to work and for his passion and help to create this fun and educational demo scenario.
– Philippe Lions, Senior Director, Product Management, Analytics Platform, Oracle Analytics Cloud (OAC) for his expertise, ideas, and help in highlighting OML working with Oracle Analytics Cloud.
– Siddesh Chikkanayakanahalli Prabhu Dev Ujjni, Staff Cloud Engineer for his dedication, hard work, knowledge, and help in creating this and other OML demo scenarios.


Source link

Leave a Reply

Your email address will not be published.