KPI Forecasting

When you create your KPI, we use advanced Machine Learning to predict future performance based upon your historic data trends and seasonality. In most cases the algorithm should be able to provide a reliable prediction, however if there have been significant changes that occurred in your historic data you will want to adjust the settings to get a better prediction.

Each setting has an effect on the prediction, however as each data set is unique you will need to determine which setting to adjust based upon your understanding.

Tuning Paramaters

Strength of Holiday Effect

Holidays are special events or periods that have an impact on the time series being analyzed. Examples of holidays include national holidays, religious observances, and school breaks. Decreasing the strength reduces the impact of prior holidays in the prediction.

Country for Holiday Data

You can select a country from the drop-down which will ensure that the algorithm takes into affect days throughout the year that might affect KPI values. This ensures that days leading up to and afterwards are considered to be anomalous and do not contribute to creating the seasonality or trend.

Seasonality Mode

Seasonality mode controls how seasonality is modeled in the time series analysis.

• Additive seasonality assumes that the seasonal component of the time series has a constant magnitude, and is added to the trend and other components of the time series. This is appropriate when the magnitude of the seasonal variation is consistent across the range of the data.

• Multiplicative seasonality assumes that the seasonal component of the time series varies in proportion to the level of the trend or other components of the time series.This is appropriate when the magnitude of the seasonal variation changes as the time series grows or shrinks

Strength of Seasonality

Seasonality strength controls the weight of seasonal components in the time series.Increasing the strength helps capture repeating patterns in the data,and reducing the strength can be useful when the seasonal patterns are weak or disrupted, such as:

1. Your data has a lot of noise or randomness.

2. There have been significant changes in the underlying dynamics of the time series that have disrupted seasonal patterns. For example, if there has been a sudden shift in consumer behavior due to a major event such as a pandemic, the seasonal patterns that were previously present may no longer be reliable indicators of future behavior.

3. Short or irregular time series data where the seasonal patterns are not well-established. For example, if you are forecasting sales for a new product that has only been on the market for a few months, the seasonal patterns may not bevery clear or consistent yet. In this case, reducing the seasonality strength can help the ML model avoid overfitting to the limited data available, and produce more accurate forecasts.

Changepoint Flexibility

Changepoint flexibility controls how the model identifies potential change points in a time series data, which is used to determine trends. A larger value allows the model to be more flexible and can detect smaller changes in the data. If too large, the model may overfit the data and identify too many small and meaningless changes.

Over fitting in machine learning refers to a situation where a model learns the training data too well, to the point that it performs poorly on new, unseen data.

A smaller value leads to a less flexible model that can only identify larger changes in the data. If too small, the model may fail to identify important changes in the data. The optimal value will vary depending on the specific dataset and often requires trial and error experimentation.

Changepoint Range

This parameter determines the proportion of the historical data will be considered for creating changepoints.
It's measured from the beginning of the time series, and a value of 0.8 means changepoints can be placed in the first 80% of the historical data.

Decreasing the range (e.g., to 0.5):

- Restricts changepoints to earlier in the time series.
- Can make the model less sensitive to recent changes.
- Might emphasize longer-term trends.

Increasing the range (e.g., to 0.9):

- Allows changepoints to be placed later in the time series.
- Can make the model more responsive to recent trends.
- Might capture more recent shifts in the data.

If you believe recent changes are crucial for your forecast, consider using a larger changepoint range. If you want to emphasize long-term trends and reduce sensitivity to recent fluctuations, use a smaller range.

Data Preprocessing

If you data has gaps you have the option for selecting a preprocessing method for KPI forecasting. It's important to consider the nature of your data and the specific KPI you're forecasting. Here's a general recommendation:

For KPI forecasting, Linear Interpolation or Spline Interpolation are often the best choices. These methods help maintain the overall trend and patterns in your data, which is crucial for accurate forecasting. Linear Interpolation works well for KPIs that change relatively steadily over time, while Spline Interpolation can capture more complex patterns and is suitable for KPIs with smoother, non-linear trends.

Avoid using No Fill, as missing data points can significantly impact forecast accuracy. Fill With Previous Value can be appropriate for KPIs that tend to remain stable for periods of time, but it may not capture changing trends effectively. Fill With Zero is generally not recommended for KPI forecasting, as it can introduce artificial patterns and skew your predictions.

‍