Classification Algorithms : Significance of Binary Categorization

Algorithms in AI are one of the 3 key pillars (Objective Setting, Data and Algorithms)

Although a broad area, AI Algorithms within AI are bucketed under Classification , Multi-Class and Regression Algorithms

Classification in particular can be understood as algorithms which enable engineers to categorize the given data sets. This in itself is again a vast area and a very vital one. It is said, that Classification Algorithms are supposed to have a role to play in every AI project which anyone would come across. Typical examples could be about categorizing different kind of customers an organization has, different type of documents to be processed etc.

Although classification can be done in multiple categories, binary classification algorithms hog most of the limelight. Any guesses ..Why??

To understand this when one categorizes data sets into 2 categories (binary), we are given to believe that 50% of the time the algorithms would be predicting the right outcome; basic premise being none of the AI algorithms can be 100% accurate. There is always an element of algorithms' accuracy.

Now, one may debate in businesses it would be prudent to say you can always classify data in only 2 buckets and hence may require to create multiple ones. What it does to the prediction accuracy in lame terms. With binary for instance we know 50% of the time it is safe to assume the prediction would be right. If we classify the data in 3 buckets, the prediction success goes down to 33%. Needless to say as one would classify data sets into more buckets, the prediction success would be further reduced. This basic math substantiates the importance of Binary Classifications.

For instance one of the common examples in classification is CHURN; where the intent is to classify the customers into categories basis their loyalty to the business. A naive way to classify this would be to have 3 buckets -> 1) Loyal 2) Churn (Voluntary) 3) Churn (Involuntary). If we analyze this further, one might debate why do we need to consider the data set for involuntary churns (which typically would be governed by another process altogether, like cancelling subscription for a customer who has not paid the bills) . So even in the above scenario, one might still be able to do better justice by just having 2 categories

There are various classification algorithms to choose from : Decision Tree, KNN, KMN, Naive Bayes etc. which we'll discuss in detail in subsequent posts