Anticipating the future is often difficult, worse when we cannot accurately predict the causes. If we know enough history, we likely will find patterns to project into the future. Based on this logic, and with modern computation, we can create highly sophisticated forecasts without knowing more than the history of the data we want to predict.
Versions of this tool have been used to effectively predict:
- Inpatient Census, Admissions, and Discharges
- ED Visits
- Leaves of Absence
- Meals
- Bed turnovers
- Staff Turnover
- Overtime
- Personal Spending
Data Requirements:
- CSV file format
- >1 year of history
- >= 1 Column of purely numerical values
- >= 1 Column of dates, with the full year (20##)
Design Principles:
- Maintenance Free
- All aspects will function until none of the dates fall before 2030
- Simple
- Only complicated when making complicated selections
- Flexible
- Can adapt data in multiple ways
- Clever
- Deduces any aspects not set by the user
Updates from Latest Version – 2020.1:
- Even more dynamic
- Nearly all calculations adjust to match length of dataset available, whereas previous versions emphasized periods longer than 1 month
- Trendlines and holidays (irregular events) checked on every pass
- Used to be calculated before finding cycles
- A “+H” beside cycles where the holidays are used
- “Trend” shown wherever a trendline is applied, and the accuracy
- Old version bundled the trend and holidays with the first cycle, overemphasizing its impact
- Baseline period
- Set a date to stop training the forecast
- Ideal for demonstrating the impact of interventions
- Each holiday is evaluated independently
- Holidays used are shown, where applied, within the “Forecast Export” table
Subscribe to get access
This application is only available to subscribers.