You learnt about the following methods that are used for time series forecasting:

- Naive method.

- Simple average method.

- Simple moving average method.

- Simple exponential smoothing technique.

- Holt’s exponential smoothing technique.

- Holt-Winters’ exponential smoothing technique.

- Auto Regressive method.

- Moving Average method.

- Autoregressive moving average method.

- Autoregressive integrated moving average method.

- Seasonal Auto Regressive integrated moving average method.

- SARIMAX.

Suppose you are solving a business problem in your organisation. The solution involves the use of time-series data. How will you choose the model to be used for that time-series data?

Industry experts have studied these models carefully and have formulated a means of deciding on the model that works well in specific situations. The thumb rule for using various models in the time series data are as follows:

- When you have data points with fewer than 10 values.

- When you have data points with more than 10 values.

In the next video, let’s understand which models can be used in case the number of data points is less than 10.

In the above video, Chiranjoy explained that when you have time-series data of fewer than 10 observations that are noisy, you should use a simple moving average method because it helps cancel out the noise to some extent. This method does not work well if the data has a seasonal component or more than 10 observations.

An example in which a simple moving average method works well is the daily forecasting of stock prices. This is because the number of observations is fewer and the data is noisy. Thus, the simple moving average method is able to predict the forecast better since it takes the variation of very few data points.

If the data points are fewer than 10 but the data is neither noisy nor has any seasonal pattern, then the Naive method works well because it will forecast the next values based on the previous values of the train data. In the case of a higher number of observations(generally more than 10), the Naive forecast tends to underpredict or overpredict the values.

Thus, the naive method works better when there are fewer than 10 observations. An example of the usage of the naive method would be in forecasting the sales of a grocery store that opened recently. In this case, the sales are not much dependent on any seasonal component, and thus, the naive method can be used here

Also, in the case of non-noisy data and seasonality with data points fewer than 10, the seasonal naive method works well. An example in which the seasonal naive method works well is forecasting the sales of a newly opened store that sells umbrellas on a monthly basis.

You can read more about the seasonal naive method here.