MMOOZZGG.com

Glossary

What is predictive analytics

In short

By definition, predictive analytics is the calculation of the probability of future events from historical data: demand, equipment failure, customer churn or late payment.

AI basics · predictive analytics

MMOOZZGG.com editorsChecked: 1 min read

For a child

Before a walk, Grandma looks at the clouds and says: “Take an umbrella, it will rain.” She has seen such clouds many times and remembers what came next. A computer can also look at what happened before and hint at what will happen tomorrow: how much ice cream people will buy on a hot day, or when the swings in the yard need fixing.

In plain words

A company collects records for years: sales by day, sensor readings, payment history. Predictive analytics finds patterns in them and carries them into the future. The model looks at what usually happened before a machine broke down or before a customer stopped buying, and notices the same signs now. The answer comes as a probability or a number: “high risk that this customer leaves”, “demand will grow on Friday”. It is like a weather forecast: not exact knowledge, but an estimate with an error margin that makes planning easier.

For business

The money here comes not from the forecast itself but from the action it triggers: buy stock on time, call a customer before they leave, replace a part before a breakdown. So a project starts with the question of who will do what with the model’s number. The data sets the cost: if the history sits in different systems and has gaps, most of the work goes into collecting and cleaning it. The risk is that the past stops repeating: a new competitor, a crisis or a change in the product range breaks the patterns, and the model has to be retrained. Quality is checked on a period the model has not seen. Industry cases with their financial effect are collected in a separate review on the site.

For a specialist

Predictive analytics combines supervised learning and time-series forecasting to estimate future values or probabilities. The tasks fall into classification (churn, default or equipment failure within a set horizon), regression (revenue, repair time) and time-series forecasting (daily demand). For tabular data, teams often use gradient boosting of trees; T. Chen and C. Guestrin (2016) described XGBoost, a scalable system with an algorithm for sparse data. For time series, teams use exponential smoothing and ARIMA models, covered in detail in the textbook “Forecasting: Principles and Practice” by R. Hyndman and G. Athanasopoulos. The main evaluation error is leakage of the future into training. So teams use cross-validation with a rolling forecasting origin, where the training set holds only observations earlier than the test point. Metrics: MAE and MAPE for numbers, ROC-AUC and recall for events.

Example

A beverage distributor builds a model from several years of sales, weather and the promotion calendar. A week ahead, it warns which stores will run out of stock in hot weather, and the logistics team delivers stock in advance.

Related terms

Questions and answers

How is predictive analytics different from regular reporting? A report shows what has already happened. A predictive model estimates what is likely to happen and gives you time to prepare.
What data do you need for forecasts? A history of events with dates over a long enough period: sales, failures, payments and requests. The more complete the records and the fewer the gaps, the more accurate the forecast.

Sources

Not sure whether your company needs this?

AI solutions map
Contact us