Diagnostic
Diagnostic analytics use historical data to answer questions about why different events have happened. While descriptive analytics use historical data to display past results, diagnostic analytics take this a step further by determining the root cause behind those results. This is the first technique that leverages machine learning to provide insights. Examples of diagnostic analytics include drilling down to focus on a particular facet of data, anomaly detection, data mining to get information from a massive set of data, and correlation analysis to pinpoint cause-and-effect relationships.
Predictive
Predictive analytics use historical data to build statistical and machine learning models to forecast what will happen in the future. This is the first type of analytics in the Analytics Maturity Model that answers questions regarding a business’s future. Techniques such as neural networks, decision trees, and regression models allow predictive analytics solutions to make recommendations on the following scenarios:
- Whether or not a customer will leave for a competitor. Customer churn models use past trends to make predictions on the risk of a customer leaving. These models can help organizations make decisions on how to preemptively maintain high-risk customers’ business.
- When to replace or repair a piece of equipment. Predictive maintenance models enable organizations that rely on machines to run their business (e.g., oil and gas companies or vending machine companies) to know when they should take proactive measures to repair or replace equipment.
- Whether or not a piece of data is fraudulent. Fraud detection models will alert organizations if they find any suspicious transaction activity.
While descriptive and diagnostic analysis can be completed using traditional BI techniques, predictive analysis requires developers to have a more specialized skillset. Along with the need to properly maintain the historical data warehouse, solutions involving predictive analytics must maintain and regularly revisit model performance to ensure that the models are making well-informed decisions. While not in scope for this book, it is important to understand the tools and mechanisms used to maintain a machine learning model’s life cycle. Azure Machine Learning and Apache Spark’s MLflow enable data scientists to train, deploy, automate, and manage their machine learning models. These technologies allow data scientists to deploy models using container-based technologies such as Kubernetes or Azure Container Instances, which can be used by applications to make batch or real-time predictions.
Leave a Reply