Now, within the Downloads Estimates section of Analytics, you can effortlessly forecast trends by choosing future date ranges to analyze both their apps and their competitors' performance.
Technical information:
Past data is displayed for both past and future date ranges. Future-only data is represented as “-” in the KPI numbers.
How Forecasted Download Estimates Are Calculated?
Data Collection
To make accurate predictions, our tool uses historical data. Specifically, it uses two years of historical data combined with data from the current date range. For example:
If you request a forecast for February 2024 to June 2024, the tool will use data from February 2022 to today.
If you request a forecast for March 2023 to June 2024, it will use data from March 2021 to today.
Time-series forecaster
The time series forecaster is designed to handle time series data and can identify various patterns in your app's historical data, such as:
Weekly trends: Patterns that repeat every week.
Yearly trends: Seasonal patterns that repeat annually.
Latest trends: Recent changes and trends over the last few months.
How the Time Series Forecaster Works
The time series forecaster uses these patterns to predict future values by decomposing the time series data into three main components:
Trend: The long-term increase or decrease in the data.
Seasonality: Regular patterns that repeat over a set period (weekly, yearly).
Holidays/Events: One-off events that can affect the data.
Prediction Process
Data Analysis: The time-series forecaster analyzes your app's past download data to detect existing patterns.
Model Training: The algorithm trains a model using the identified patterns.
Forecast Generation: The trained model forecasts future download estimates by projecting the identified patterns forward into the requested date range.
Why Predictions Sometimes Cannot Be Made
There are instances where our tool cannot provide accurate forecasts. This can happen due to:
Insufficient Historical Data: If there isn't enough historical data, the algorithm cannot identify reliable patterns.
Erratic Data: If the historical data shows highly irregular fluctuations without a clear pattern, the time series forecaster may not be able to make reliable predictions.
Sudden Changes: Significant and sudden changes in user behavior, market conditions, or external factors that do not follow past patterns can lead to unreliable forecasts.
Common Issues
Non-repetitive Patterns: If your download data does not have repeating trends or shows random spikes and drops, the time series forecaster may indicate it cannot make a forecast.
Data Anomalies: Sudden and unexplained anomalies in the data can disrupt the pattern recognition process.