Date that article was published: October 25, 2018

Summary of Article

Benford’s law is a statistical law about the distribution of first/significant digits, which has been utilized for anomaly and fraud detection. More specifically, it states that numbers found in a series of records usually do not display a uniform distribution but instead display a distribution that is tail-heavy distribution such that the digit “1” is the most frequent followed by the remaining digits in numerical order. In general, numerical records that follow Benford’s law represent magnitudes of events, have no pre-established minimums and maximums, are not made up of numbers used as identifiers (ie. SSN, phone numbers etc.), and have a mean less than the median.

drawing

Areas of Application

Benford’s law has great application in catching anomalies or fraud detection. In the business world, this could mean finding ‘manufactured’ data in financial statements of a business. Additionally, this could mean detecting fradulent activity in a business’ network traffic. However, it is important to keep in mind that Benford’s law is simply a tool that shouldn’t be used to make the final decision. Its main purpose should be for initial screening.

What Do I Think?

I think the article explores a very interesting topic. I wanted to delve more into this topic as it is something I worked very close with at my internship last summer. In fact, I used Benford’s analytic data to detect anomalies in the network traffic of the client I was working for and was actually able to generate a potential observation for an anomaly.

Author Information

Tirthajyoti Sarkar is the author who is an open-source contributor specializing in AI and Data science. His interests include AI, analytics, and Industry 4.0. He is also a Medium member since August 2018 and the Editor of Productive Data science and writes about topics similar to Benford’s law Analytic. More articles by him can be found below.

Article highlights (what I found useful):

The article does a very good job in explaining the intricacies of Benford’s law in a way that is understandable to both technical and non-technical audiences. As someone who has more of a technical background, I found the breakdown of detecting fraud in a regression model very useful. Likewise, the section about business applications changes to a broader scope, making it understandable to non-technical audiences.