Time-Adjusted Item Request Prediction for Low-Activity Events
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing data prediction technologies fail to capture recent trends or changes that impact future events, leading to inaccurate predictions, especially for low-activity events, and result in inefficient resource consumption due to under-predictions or over-predictions.
Innovation Solution
A data prediction subsystem that uses a triple moving average and improved rounding process to determine accurate and reliable predictions, transforming non-integer values into interpretable integers, and accounts for location-specific and time-dependent event patterns.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If previous data prediction technology bases predictions on events from the same time period in the previous year, then predictions can be generated using available historical data, but recent trends and changes are not captured leading to inaccurate predictions
Solution Approach 1:
The patent implements dynamic prediction by adjusting the time window for data collection based on event frequency. For high-frequency events, it uses a rolling 12-month window that continuously incorporates recent data while excluding outdated information. This dynamic approach ensures the prediction model adapts to recent trends and changes rather than relying on static historical comparisons from the same period in previous years.
Solution Approach 2:
The system performs preliminary data filtering and time-window adjustment before generating predictions. By pre-processing the historical data to identify and exclude periods with insufficient activity (below threshold event frequencies), the system ensures that only relevant, information-rich periods contribute to the prediction, preventing loss of recent trend information.
2Productivity
If data prediction technology uses limited available information for prediction, then predictions can be made with available data, but reliability is reduced when large amounts of information are not available
Solution Approach 1:
The patent dynamically changes the parameter of time window duration based on event frequency characteristics. For low-frequency events where sufficient data may not be available in standard windows, the system adjusts the window size and weighting schemes to maximize the use of available information while maintaining statistical reliability. This adaptive parameter adjustment allows reliable predictions even when large amounts of historical data are not available.
Solution Approach 2:
The system applies partial weighting to historical data points based on their relevance and sufficiency. Rather than requiring complete information from all time periods, it selectively incorporates data from periods where event frequency meets thresholds, using partial data from some periods and excluding others. This approach enables predictions to be made with the partial information that is available while maintaining reliability through selective data inclusion.
3Adaptability or versatility
If data prediction technology predicts events with low activity frequency, then coverage is improved, but prediction accuracy decreases due to insufficient data
Solution Approach 1:
The patent merges data from multiple sources and time windows to improve predictions for low-activity events. By combining information from location-specific event data with zone-level aggregated data and item-type aggregated data, the system creates a multi-layered prediction model that provides sufficient statistical basis even for events with low frequency at individual locations. This merging approach maintains accuracy while extending coverage to low-activity events.
Solution Approach 2:
The system adds dimensional aggregation by combining data across multiple dimensions: location-specific data, zone-level data, and item-type data. For low-activity events at a specific location, the prediction model incorporates data from the same item type across different locations and from the same location for different item types, effectively moving from a single-dimension to multi-dimensional data analysis that improves accuracy for rare events.
4Loss of energy
If proactive request system under-predicts item needs, then resource consumption is reduced, but supplemental requests are needed resulting in wasted communication and computing resources
Solution Approach 1:
The patent implements feedback mechanisms where prediction accuracy is continuously monitored and the model is adjusted based on actual event outcomes. By comparing predicted item needs with actual consumption patterns, the system learns from discrepancies and refines its predictions, reducing both under-prediction and over-prediction. This feedback loop ensures that resource consumption is optimized while maintaining accurate item need information.
Solution Approach 2:
The system performs preliminary prediction with confidence intervals and uncertainty analysis before generating proactive requests. By evaluating the reliability of predictions and incorporating safety margins based on data quality and event frequency, the system determines the appropriate request quantity in advance, reducing the need for supplemental requests while avoiding excessive resource consumption.
5Reliability
If proactive request system over-predicts item needs, then item availability is ensured, but wasted resources occur in transporting unneeded items
Solution Approach 1:
The patent dynamically adjusts the safety margin parameter in predictions based on event frequency and data quality. For high-frequency events with reliable historical data, smaller safety margins are applied, reducing over-prediction. For low-frequency events or periods with limited data, larger safety margins are used to ensure availability. This adaptive parameter adjustment balances reliability with resource efficiency.
Solution Approach 2:
The system applies partial over-prediction selectively rather than uniformly across all events. By using confidence-based prediction adjustments, it only incorporates safety margins when necessary to ensure availability, avoiding excessive transportation of unneeded items for events where predictions are already highly reliable. This partial application of over-prediction maintains item availability while minimizing waste.
Data Source
AI summary
A data prediction subsystem stores hourly event potential data indicating an expected amount of removal events as a function of time of day. Based at least in part on event data, it is determined that a first item associated with a first location has an empty status at a time after a start of a day. For the day, an anticipated event value is determined for the first item at the first location. Using the anticipated event value and the hourly event potential data, a time-adjusted event value is determined. Based at least in part on the time-adjusted event value, a prediction value is determined that corresponds to a recommended amount of the first item to request for a future time.