# Forecast **Forecast** is a long-range capacity planning feature that analyzes historical performance data to project future trends over a **180-day rolling window**. It helps organizations proactively manage infrastructure by identifying when to **scale up or down resources** such as CPU, memory, disk, and **throughput**. ## **How Forecast Is Created** **Data Collection** * A minimum of 14 days of historical performance data is required. * 180 days is the standard data window for training the forecast model. **Model Generation** * Forecasting models are generated for each metric. * Models are refreshed weekly (every Sunday) with a moving 180-day window. **Model Type** * Machine Learning models for time-series forecasting. * Focused on performance metrics only (not availability metrics). * Each model learns the behavior pattern and forecasts based on statistical trends. ## **How it works** Infraon’s Forecast uses intelligent algorithms to study past data and predict future trends. It looks at how a metric (like CPU or bandwidth usage) has behaved over time and tries to spot patterns. The forecast shows a range using upper and lower lines (called bands) to give you a safe estimate of how things might go. The middle line is the expected value. You can also fine-tune forecasts using settings added in the terminologies below. **Terminologies |** Forecast Details


Label	Action	Description / Example
Trend Analysis	Detects the overall direction of usage	Helps identify if resource usage is increasing, decreasing, or stable over time.
Seasonality	Finds repeating usage patterns	Spot regular patterns like higher usage during weekdays or business hours.
Change Point Detection	Flags sudden shifts in behavior	Detects when there is a significant change in the trend (e.g., after a software update).
Confidence Bands	Shows a safe prediction range	The forecast graph includes a range (upper & lower bands) with a center line for expected use.
Predicted Value	Calculates expected usage	The central point between the upper and lower bands is used for planning.
Change Point Scale	Adjusts sensitivity to changes	Higher value = model closely follows recent changes; lower = smoother trends.
Include History	Shows past data alongside forecast	Useful to compare how the system performed before and how it might behave going forward.

## **What do you see on the screen** When you open the Forecast view, you'll see a line graph showing past and predicted values for a selected performance metric. The graph helps visualize how usage has behaved historically and how it’s expected to trend. Hovering over any point in the graph displays detailed forecast data for that timestamp. **Basic Details |** Forecast


Label	Action	Description / Example
Polled	Displays past polled data	This is collected (raw) data from devices at periodic intervals (e.g., every hour).
Predicted	Displays forecast trend	Predicted future usage based on past patterns.
Grey Area	Shows forecast range	It covers the lower band to the upper band, representing the confidence range.
Upper Band	Forecast upper limit	Maximum possible usage at a given time based on model predictions.
Lower Band	Forecast lower limit	The minimum possible usage forecasted by the model.
Hover Preview	Shows details at a specific timestamp	View time, polled data, and forecast values (Upper, Predicted, Lower). Example: Time: 18th May, 2025 07:00 PM Polled Data: 5.50 % Upper Band: 6.47 % Predicted: 6.07 % Lower Band: 5.57 %
Filter
Current Month	Filter forecast to the current month only	From today's date to the end of the current month.
Next Month	Show forecast for the next whole month	For example, from June 1 to June 30.
Next 3 Months	View medium-term trend	Displays forecast for 90 days.
Next 6 Months	Show long-term usage prediction	Forecasts for up to 180 days into the future (maximum window).

### **Test Forecast** The **Test Forecast** feature allows users to generate a **real-time forecast** by temporarily rebuilding the model using **customized settings**. This is especially useful when users want the estimates to reflect unique scenarios—such as holidays, maintenance windows, or unusual business cycles—where typical patterns may not apply. **Test Forecast |** Basic Details


Label	Action	Description / Example
Business Hour Profile	Forecast only during business hours	Example: Monday–Friday, 09:00–18:00. Excludes weekends or custom time blocks.
Change Point Scale	Set model sensitivity	Range: 0.01 = underfit, 0.5 = overfit, 0.1 = balanced.
Include History	Toggle past data in the graph	Enables visual comparison of historical and forecast data.
Weekly Seasonal	Enable weekly pattern recognition	Helpful in detecting trends like high CPU usage every Monday.
Daily Seasonal	Enable intra-day pattern detection	Identifies time-of-day peaks, like lunch-hour bandwidth spikes.

### **Polled Data** This table displays the historical values collected from the device or component at regular polling intervals. It helps users understand how the resource has performed over time.


Label	Description / Example
Time	The exact timestamp when the data was collected (e.g., May 01, 2025 12:00 AM).
Value	The actual metric value now (e.g., 7.63%).

### **Forecasted Data** This table displays the **predicted future values** for the selected metric. The forecasting model generates these values using historical trends and helps anticipate upcoming changes in resource usage.


Label	Description / Example
Time	Future timestamp for calculating the forecast (e.g., May 01, 2025 01:00 AM).
Upper Band	The highest possible predicted value based on trend analysis (e.g., 8.03%).
Predicted	The most likely expected value (center of the forecast range) (e.g., 7.64%).
Lower Band	The lowest possible predicted value based on trend analysis (e.g., 7.20%).

{% hint style="info" %} **Export** The forecast graph can be downloaded as a **PDF**, while the **Polled Data** and **Forecasted Data** tables can be exported in **CSV format** for detailed offline review or further processing. {% endhint %}