Exponential Smoothing Exponential smoothing is a time series forecasting method. There are three main types, simp

150 words to each questions

Q1. Exponential Smoothing

Exponential smoothing is a time series forecasting method. There are three main types, simple, double, and triple exponential smoothing (Brownlee, 2020). In exponential smoothing models are weighted sums of past observations where older data weights less than newer one.

              Figure 1 is an example of exponential smoothing in action.

Figure 1. Exponential Smoothing. (Vandeput, 2019).

In Figure 1 we can observer the blue, jagged line, representing demand over a period. An untrained eye can pick the upward trend, however, if we were to shrink the time interval this could be harder to notice. Additionally, there are data sets which true patterns are harder to understand. With exponential smoothing, represented by the orange line in Figure 1, an analyst can see the underlying trend. In figure 1 this trend is upwards, signifying that the demand has increased over time. 

              As mentioned before, exponential smoothing can account for seasonality. According to Brownlee (2020), to account for this seasonality trends analysts must employee the following variables: alpha, beta, trend type, demand type, and phi. Alpha is the smoothing factor for the level. Beta is the smoothing factor for the trend. Trend type can be additive or multiplicative. Dampen type can be either additive or multiplicative. Finally, phi, the damping coefficient.  

              According to Brownlee (2020), other variables can also aid the analyst produce a more accurate model. These include gamma and the period. Gamma is a smoothing factor for the seasonality. The period is time steps in the season. 

              Understanding these variables is crucial in the planning and creating of an exponential smoothing model. Although the arithmetic calculations can be rigorous, data analysts rely on the use of statistical software’s perform later. 

Bibliography

Brownlee, J. (2020, April 12). A Gentle Introduction to Exponential Smoothing for Time Series Forecasting in Python. Retrieved from Machine Learning Mastery: https://machinelearningmastery.com/exponential-smoothing-for-time-series-forecasting-in-python/

Vandeput, N. (2019, Novemeber 12). Simple Exponential Smoothing for Time Series Forecasting . Retrieved from Towards Data Science: https://towardsdatascience.com/simple-exponential-smoothing-749fc5631bed

Q2.K-mean (k-mean) clustering is a technique of plotting specific items that show correlation within groups or clusters and measuring the distance from a specific center point of the cluster.  The center point is referred to as a centroid(s).  Clustering is an unsupervised technique to group items or objects that have similar attributes (EMC Education Services, 2015, p. 153).  These clusters generally identify hidden attributes to data that may not be normally viewed during data manipulation activities.  While developing k-means, the first step is to estimate the centroids, in an attempt to locate the center or best fit (this may be hard to locate and may not be perfect in the end).  The clusters with centroids are calculated for best fit, and recalculated to make sure.

            One of the best examples about k-means clustering is that of grocery store design.  From information from shoppers and the habits in buying like items (linked) clusters in data can be visualized.  Once visualized a grocery store has two decisions to make cluster like items together so that consumers buy items in clusters or they may try to design a store that links many clusters together so that consumers buy items from different clusters as well as items they my not need but they associate them together with the item that was needed.  

Example one: A store that lumps the milk isle, cereal isle, and cookie isle all together because these three items may be frequently purchased together. 

            Example two: A store that alternates clusters: milk isle, canned soup isle, cereal isle, bread isle, cookie isle, baking isle.  So that different cluster are hit on each isle, but alternating clusters so more product is bought.

            These examples come from two different stores locally in Fairbanks, AK. 

The information collected from shopper habits has been instrument in the development of grocery stores and how they are laid out.  These strategies also allow for room for season items normally found at the main entrance or exit.  Where again “Fall” items are clustered together to promote multiple item purchases.

References

EMC Education Services. (2015). Data Science & Big Data Analytics: Discovering, Analyzing, Visualizing and Presenting Data. Indianapolis, IN: John Wiley & Sons, Inc.