Improve Anomaly Detection to Better Predict Future Demand

Using anomaly detection in time series forecasting 

Anomaly detection – sometimes referred to as outlier analysis –  identifies data points or ranges that deviate from the dataset’s normal behavior. Understanding what causes  anomalous data is critical for forecasters to detect future problems or opportunities for their business. 

Time series forecasting uses statistical approaches and machine learning (ML) techniques to predict future demand based on historical data. Teams use various anomaly detection techniques to accurately analyze time series data to identify irregularities and improve overall prediction accuracy. 

But whether you’re using clustering-based anomaly detection or density-based techniques to identify time series outliers, it’s imperative to use high-quality, accurate datasets when performing anomaly detection. 

The value of high quality data in anomaly detection

Forecasting experts identify data patterns that are normal using machine learning models. Doing so requires a tremendous amount of accurate data around trend and seasonality. But what about the external data that isn’t always accounted for? Often forecasting teams have access to historical business trends and seasonality data, but it’s challenging to find forecast-grade external data sources that will integrate into ML models. 

Time series forecasting helps teams across the company prepare for future demand needs by estimating them with current data. But if forecasters don’t have the full picture of the external circumstances influencing demand and causing data anomalies, it becomes impossible to predict what’s coming.

Demand intelligence explains data anomalies in your time series data

Demand intelligence is advanced insight into what is causing anomalies in time series data. By aggregating demand causal factors such as scheduled events, health warnings, academic events and more, demand forecasters can improve anomaly detection efforts and train their time series forecasting models with this new dataset.   

With PredictHQ’s demand intelligence API, data teams can access millions of verified demand causal factors that cause data anomalies. We’ve included features in our data to assist your team in discovering what’s driving demand anomalies and prioritize the most impactful catalysts as you incorporate these into your ML models. 

• Ranking: Time series models need to know the demand causal factors to prioritize so PredictHQ created three proprietary ranking algorithms to identify the impact of events. These are optimized to highlight impact based on attendance, location, and anticipated air travel.  

• Event categories: PredictHQ covers 19 event categories globally so you can identify the types of events that catapult anomalies in historic data sets. 

• Aggregate Event Impact: Understand impact at an aggregate level vs individual event-level. 

Detecting anomalies faster leads to accurate demand prediction

Anomaly detection at scale is essential for forecasting accuracy. Without the proper data and tools, anomaly analysis and detection can be a daunting task for data teams. Our automated relevancy engine automatically detects anomalies in your transactional data based on reconstructed demand patterns. Now you can reveal the factors driving demand and increase the speed of anomaly detection so you can always be prepared for them.