Time-Series Analysis Using Moving Average in WEKA

Time-Series Analysis with Moving Average in WEKA is usually done using the time series forecasting package that extends WEKA's functionality. The Moving Average (MA) technique helps smooth time-series data and can also be used for prediction.

1. Install the Time Series Forecasting Package

• Open WEKA GUI Chooser → Tools → Package Manager.
• Search for timeSeriesForecasting (WEKA package).
• Install and restart WEKA.

2. Load Time-Series Data

• Format: .arff or .csv.
• Example (monthly sales data):

@relation sales
@attribute month date "yyyy-MM"
@attribute sales numeric

@data
2020-01, 120
2020-02, 135
2020-03, 128
2020-04, 140
2020-05, 150

The date/time attribute is required.

3. Open the Forecasting Environment

• From WEKA Explorer → Tools → Time Series Forecasting.

4. Configure Forecasting

• Select the target variable (e.g., sales).
• Choose the lag creation (lags are past values used for prediction).
• In Base Forecaster, choose:
  • weka.classifiers.timeseries.core.TSLagMaker (handles time-based lags).
  • For moving average, you can use:
    • weka.classifiers.functions.SimpleLinearRegression (with lagged values).
    • Or specify moving average window under lag settings.

5. Moving Average Setup

• In LagMaker, set time lag length (e.g., 3 for 3-month moving average).
• WEKA will then create new attributes like:
  • sales_lag1 (previous month),
  • sales_lag2, sales_lag3.
• The model will use these lagged values to compute a moving average forecast.

6. Run the Forecast

• Click Start forecasting.
• WEKA will output predictions for future time steps along with actual values.