W11_UDS_Forecasting Method Comparison

1. Problem Evaluation

Sometime we face problem when we have to predict future condition. This prediction is very important to us in making decision and choosing strategy for company. Forecast methods are usually used to predict future condition.

There are two type of forecast method:

1. Time series models; What happened in the past will happen again in the future (with consistent error)
2. Regression Analysis; One variable is affected by other variables

 

2. Development of feasible alternatives

This small research will evaluate forecast method focusing in time series models. Time series model are common model in forecasting because user friendly use and useful in determine future condition. There are three Time series models that will compare:

1. Simple Moving Average
2. Weighted moving average
3. Exponential smoothing

 

3. Development the outcome for each alternative

This time series model comparison will evaluate using two forecast error measurements: Mean Squared Error (MSE) and Mean Absolute Deviation (MAD). Using MSE and MAD approach we can get difference between actual data and forecast calculation. Comparing the value of MSE and MAD value each model are the good way to know which the best alternative is.

4. Selection of criteria

The best model is the model that has the lowest error. Error formula is actual data minus forecast result. Based on those criteria, we will choose model that have lowest MSE and MAD. 

5. Analysis and comparison of the alternative

To simulate this comparison we will use one of the product demand data. Those data consist of 70 periods.

Table 1. Alternatives data

Next step is data calculating forecast prediction using three methods formula (Simple Moving Average, Weighted moving average, and Exponential smoothing).

Equation 1. Simple Moving Average, Weighted moving average, and Exponential smoothing Formula

The Calculation result of three forecasting methods are in table below:

Table 2. Simple Moving Average calculation result

Table 3. Weighted Moving Average calculation result

Table 4. Exponential smoothing calculation result

Table 5. Comparison Result

6. Alternative selection

As can be seen above there is no dominant model in MSE and MAD calculation result. The lowest result value of MSE is Exponential Smoothing model with alpha 0,5 and the lowest result value of MAD is three period Moving Average model.

Based on analysis above there is no the best model of forecasting, but we can conclude the appropriate model of each variances model (lowest error index in same variance). The appropriate model of each variances are list below :

1. 3 period moving average
2. 135 weighted moving average
3. Exponential smoothing with alpha 0.5
   

7. Performance monitoring & Post Evaluation Result

To get the best result of forecasting we have to find most appropriate model of forecasting. We should check the error index first (MSA and MAD) of our historical data. When you find the lowest error index in some forecast model, it should be the most appropriate model of your data. We should do the step above because every data have different characteristic.

References

1. Harvard Business Review (2017). How to Choose the Right Forecasting Technique. Retrieved from https://hbr.org/1971/07/how-to-choose-the-right-forecasting-technique
2. com (2017). JD Edwards EnterpriseOne Applications Forecast Management Implementation Guide. Retrieved from https://docs.oracle.com/cd/E16582_01/doc.91/e15111/und_forecast_levels_methods.htm#EOAFM00177
3. Data Science Central (2017). Selecting Forecasting Methods in Data Science. Retrieved from https://www.datasciencecentral.com/profiles/blogs/selecting-forecasting-methods-in-data-science