Machine learning can be used to solve a bunch of problems that are in general hard to solve. The kinds of problems machine learning can be used to solve are, as we discussed earlier, highly varied. When Spotify recommends you to add W.A.P. to your morning playlist, and when Gmail marks the email from the Nigerian prince as “Spam”, what goes behind the scenes is completely different because they have very separate algorithms. We can classify Machine Learning algorithms based on what kinds of problems they solve. Here are some of the problems which are the most commonly solved problems :

Imagine that you’re in the mood to watch a good movie, but you have no clue whatsoever about what to watch. So you sit down on your soft brown sofa, pop a can of coca cola, and start scrolling through the home page of Netflix. Voila, your eyes wander to the cover of the movie Shrek. Yay!! Time to binge on all the Shrek movies, and then onto the Pirates of the Caribbean. You will not regret watching these movies all night, and your stomach will thank you endlessly for all the cheese puffs you shoved down your throat. Of course, you can always blame Netflix’s recommendation algorithm for this, and avoid hurting your ego. These are the algorithms that use your viewing history, along with the viewing history of other people histories, to make sure that you can always find a movie that will make you spend all night on Netflix. Matrix Factorisation, Alternating Least Squares, User-based K-Nearest Neighbours are some examples of recommendation algorithms.

And whenever you search for a term on Google or a product on Amazon, you get a long list to choose from. But if you’re lucky, you’ll find your favourite item in the beginning on the list, and almost always on the first page of results. When do you last remember going to page 2 of Google to find what you need? Probably a long time ago. This is due to Google’s very famous ranking algorithm. Ranking algorithms sort search results based on various factors, in this case it could be hits, relevance, domain age, user friendliness and the quality of the content. Now you must be wondering why our website isn’t the top result when we search for “Machine Learning Algorithms”. Some of the most commonly used ranking algorithms are LambdaRank, RankNet, and LambdaMART.

Now let’s come back to when you were browsing Amazon, looking for let’s say a ceiling fan. You pick your favourite brand and color, and click “add to cart”. As you do that, you see other products, a rope for example, under the section “customers also bought”. This section works thanks to a clustering algorithm where the model understands that some entities, products in this case, are more related to some than the others. Other examples where clustering algorithms are used are YouTube’s Mixes and Spotify Radio. Affinity propagation, agglomerative clustering, BIRCH and DBSCAN are some of the well known clustering methods.

Remember that time when you got an email from the Nigerian prince, who wanted to recognise you for all the hard work you had been doing, and wanted to give you 1 million dollars, except that because of the way his bank account was setup you had to send him 1000 dollars first and then you would get your million dollars? Remember how gmail was so greedy that it put the email in spam, and probably sent him 1000 dollars and took your million? Well you can thank classification algorithms for stealing the chance to buy a matte black Lamborghini. These are algorithms that are used to take a bunch of things, could be emails, images, or texts, and classify them into different groups. Examples of common classification algorithms are Decision Tree, Random Forest, Naïve Bayes, etc.

Then let’s say you just casually start browsing Twitter because you suddenly felt woke and wanted to get your opinions validated from other woke people on the internet, and then half an hour later you regret wasting so much of your time getting roasted by some feminist society of Tanzania. So to feel better, you switch to YouTube to waste more of your time, because come on, we all know no one stops at just one video. After you’ve learnt everything there is to know about viral street pranks, anime conspiracies and 5-minute DIY life hacks, you realise that there was one thing that was common in the two cases - the “Trending” section. How this works is, it looks for topics and keywords whose trends have grown rapidly in a short span of time. This is a simple example of anomaly detection through machine learning. It is a way to find anomalies in a dataset, points that don’t follow the common trend. It is applicable in a variety of domains, such as intrusion detection, fraud detection, security and surveillance. Outlier analysis, Bayesian networks, and self-organised maps are some algorithms used for Anomaly Detection.

We’ve all had that one day, when the school’s principal came into our class and said that our class was the warasht worst class the school ever came across. I think it’s very obvious to all of us that she used a regression model to come to that conclusion. She probably compared all the bad events with a scale of how bad they were, and compared them with the value of the percentage of the culprits that were in your class, and saw a clear correlation. Regression algorithms are used to find correlations between different variables, for example, if a company wants to know the effectiveness of their marketing, they would try to find a correlation between the money they spend on marketing vs their revenue. This is very useful, as this can help pinpoint problems, inefficiencies, and areas to improve in. Some exmaples of Regression algorithms are linear regression, multivariate regression, and support vector machines.

Note: This article is not sponsored by Netflix, Coca Cola, Amazon, Spotify or Turing. Although, if you wanna pay us, we don’t mind.