Machine Learning vs Statistical Modeling

Hello! In this video, we’ll be covering the differences
between machine learning and statistical modeling. Statistical Modelling and Machine Learning
can be mixed up sometimes. So, to clarify …
Machine learning is an algorithm that can learn from data without being reliant on standard
programming practices, like Object Orientated Design. Here are some important facts about Machine Learning
* Machine Learning is a newer field of study than statistics (for instance, Machine Learning was invented
in 1959, whereas statistics originated in the 17thcentury)
* Machine Learning can result in more detailed information than statistical modelling. * Machine Learning is a subfield of computer
science and A.I., and contributes to building systems that can learn from data without explicit
programming * Finally, Machine Learning uses fewer assumptions
than statistical modelling Statistical Modeling is the formalization
of relationships between variables in the form of mathematical equations. Statistical Modelling is a subfield of math
that deals with finding relationships between variables to predict outcomes. It deals with a small amount of data with
fewer attributes and, as such, there is a good chance that over-fitting will occur. Statistical Modeling requires the modeller
to understand the relation and implementation that a variable has on an equation, in an
effort to best ‘estimate’ the function output to a certain error. In comparison, machine Learning requires minimal
human effort, as the workload involved in computing is placed squarely on the machine. Furthermore, Machine Learning has a strong
predictive power, as the machine is ‘fit’ and ‘trained’ to find patterns in the data. Here’s a table that details the different
naming terminologies between machine learning and statistical modeling. Please take a moment to review the chart
Beyond naming convention, there are several other differences between machine learning
and statistical modelling. This chart summarizes a few of them. [1] For instance, in machine learning, fewer
assumptions are made, due to a better accuracy from the predictive models, in comparison
to statistical modelling which is more mathematically based. [2] Machine Learning is a subfield of Computer
Science and uses algorithms, while Statistical Modelling is a subfield of Mathematics and
uses equations. [3] One of the main things that makes machine
learning useful is that it also works well with large sets of data, whereas statistical
modelling has a hard time doing so. Machine learning provides strong predictive
ability with minimal human effort, while statistical modelling provides the best estimate and more
human effort. So? how does Machine Learning actually work? Well, one of the more important concepts to
know in Machine Learning is being able to distinguish supervised and unsupervised learning. In a later module, we’ll cover supervised and
unsupervised learning in more depth, but for now here is a brief synopsis:
In supervised learning, we have a set of training data, or labeled data, in which we know the
structure and the outcome of it. We take this data and train a machine learning
model, so it can understand patterns in the data. Once the model has been trained, we can use
it to predict the results of out-of-sample data, or data in which the results are unknown. Conversely, if we are given a set of data
that is unstructured, then we can apply unsupervised machine learning models to find patterns that
exist within that data. Thanks for watching!

8 thoughts on “Machine Learning vs Statistical Modeling

  • I don't agree.
    Statistical modelling AND machine learning are looking for relationships between variables.
    There are many assumptions one must make when using machine learning techniques and you must choose the correct model.
    A statistical model such as lm can be used on 'big data' and can be just as predictive as a machine learning algorithm depending on the data.

  • This is an, if not erroneous, incomplete presentation of machine learning vs statistical modeling. Designed experiments is a vital part, if not a more complete form of statistical modelling than predictive statistical modeling. It allows you to model cause and effects and establish a base for mechanistic modeling. If you do not know why you get the prediction you get from the ML algorithms, you can "drill down" with a set of designed experiments. A/B testing is primitive but very commonly used designed experiment e.g. in web design. Thus, it is reasonable to assume that the most productive development environments will include both the machine learning toolset and the statistical modelling toolset. This especially if you are working in a cross-disciplinarian environment such as the auto, power or aerospace industries. Environments that also frequently use first principle simulation tools. You may assume that all these tools will merge into a powerful integrated toolset. Or?

  • This video is completely misinforming the viewer. Machine Learning is solidly based in statistics, but contextualizes the objectives in a descriptive rather than inferential way.

  • Why the output of a standard regression (with p-values, coefficients, etcetera) doesn’t look like a machine learning regression output?

  • Why the output of a standard regression (with p-values, coefficients, etcetera) doesn’t look like a machine learning regression output?

  • Are you kidding me? Algorithms based on mathematics too. Based on this video you know nothing about machine learning or even computer science itself. Lol

Leave a Reply

Your email address will not be published. Required fields are marked *