Monthly Prediction of the US retail and Food Services sales Published Monthly By the US Census Bureau with an Accuracy of ~97.76% Average

The Business & Economy topic is your portal to comprehensive information about the number, activity, characteristics, and locations of American businesses.

The Census Bureau has been collecting information about American businesses for over 200 years. Today, statistics on almost 27 million business locations are among the Nation’s most important. Business statistics are used to help start a business, compare industries and locations, assess the economic health of local communities, or just to make informed decisions.

What is a Predictive Analysis?

Predictive analysis is the process of making predictions based on past and present data and analyzing hidden trends or patterns in the data. In other words, it’s a method of discovering future events or predicting future sales before it happens. This is not a new concept, but with the help and popularity of Python, a high-level interpreted general-purpose language, and its rich host of numerous libraries, such as Sklearn, Pandas, Matplotlib, Numpy and others, anyone fascinated with Machine Learning and Artificial Intelligence (AI) can develop a predictive model.

A practical example is a meteorologist who predicts weather by forecasting expected weather outcomes based on past and present data by analyzing trends using a predictive model. However, this might not be 100% accurate, but with a high accuracy rate. Let’s look into some practical areas where I have tested some predicted models developed using Python, Sklearn, Mathplotlib, Pandas, Numpy, Seaborn, and other libraries.

I’m very fascinated with technology and have spent a great amount of time studying and practicing the machine learning concept. However, it’s equally important to put some of that self-taught knowledge into a real-life problem to determine how well this model can perform in production. There are various types of machine learning models, depending on the problem one would like to solve – this will determine the type of machine learning model to solve that specific problem.

Predictive models can be used in the health sector to predict the likelihood of someone having a heart attack, by selecting appropriate features to achieve a high performance. Irrelevant or partially relevant features can negatively impact model performance. For binary outcome (Yes/No output), a Logistic Regression model will be appropriate using given factors to determine if a patient will likely have a heart attack or not.

In the banking sector, predictive models can be used to predict someone that would most likely default on credit card payment. Predictive models are used in the bank sector to protect banks from issuing a credit card to someone that their model deemed most likely to default on repayment. The business sector uses predictive models to determine future sales and other future events, such as the number of returning customers or revenue forecast. In the process of developing a prediction model, there are critical steps necessary to get optimal performance. Machine Learning algorithms and Artificial intelligence lifeline is data, the more accurate and refined data available the more optimal performance one can get from machine learning algorithms. However, in real life, perfect data is impossible, as most data scientists and machine learning engineers spend most of their time in data wrangling or data cleaning before that data can be useful and ready for the training and testing process of the model. Overfitting and underfitting are also very important to avoid bias in trained model.

How does Machine Learning Model Works?

To develop a machine learning model, there must be data from the previous activities or transactions data. Those data will be used to training/test and fit the model to predict future results, if the model is given new data that it has not seen before. When new data is present to the model to predict on those new data, the trained model will perform prediction based on the previously train, test, and fit method.

A Practical Example of How Machine Learning Algorithms Were Tested Against Future Data

I developed a machine learning model in April 2020 to predict the monthly U.S retail sales published monthly by the U.S Census Bureau. The overall performance of this model has been an average of 95.4% accuracy, which is pretty good in predicting the real-life U.S monthly sales. I compared the actual figures from the US monthly retail sales to my trained model and the results were very impressive. I have been testing this model every month and overall performance seems astonishing, see the graph in the next page.

• Report posted monthly by the U.S Census Bureau between 15 & 17 each month
• Retail sales report a strong indicator of economy health determine if contracting or expanding
• Retail sales in terms of direct economic activity account for almost one-third of GDP
• Used by the Labor and Statistics Bureau to calculate Consumer Price Indices and productivity measurements

Month	Actual	Prediction	Diff (Pred – Act)	Percentage Accuracy
April 2020	403,946	398,821	-5,125	98.73%
May 2020	485,545	382,152	-103,393	78.71%
June 2020	524,306	456,695	-67,611	87.10%
July 2020	535,982	528,345	-7,637	98.58%
August 2020	537,526	525,885	-11,641	97.83%
September 2020	549,256	535,695	-13,561	97.53%
October 2020	553,329	481,220	-72,109	86.97%
November 2020	546,504	585,088	38,584	107.06%
December 2020	540,916	563,475	22,559	104.17%
January 2021	568,215	529,616	-38,599	93.21%
February 2021	561,685	568,891	7,206	101.28%
March 2021	619,105	563,898	-55,207	91.08%
April 2021	619,923	619,556	-367	99.94%
May 2021	620,214	620,088	-126	99.98%
June 2021	621,340	606,525	-14,815	97.62%
July 2021	617,719	624,297	6,578	101.06%
August 2021	618,676	619,310	634	100.10%
September 2021	625,416	620,271	-5,145	99.18%
October 2021	638,189	627,038	-11,151	98.25%
November 2021	639,829	639,863	34	100.01%
December 2021	626,833	641,510	14,677	102.34%
January 2022	649,776	628,418	-21,358	96.71%
February 2022	658,128	651,455	-6,673	98.99%
March 2022	665,726	659,841	-5,885	99.12%
April 2022	677,711	667,470	-10,241	98.49%
May 2022	672,874	679,504	6,630	100.99%
June 2022	680,591	668,366	-12,225	98.20%
July 2022	682,815	682,395	-420	99.94%
August 2022	683,291	684,628	1,337	100.20%
September 2022	683,974	680,317	-3,657	99.47%
October 2022	694,518	685,792	-8,726	98.74%
November 2022	693,482	696,379	2,897	100.42%
December 2022	677,136	695,339	18,203	102.69%
January 2023