What do you mean, Machine Learning Predictive Model could be used even for this?

6 Jun 2022

So, where exactly is machine learning used?

If you’re unfamiliar with the AI(Artificial Intelligence) industry, you are bound to show a puzzled face when discussing AI machine learning solutions. As one familiar reference to the list, AI models have been appearing from here and there on commercials. Rosie, who has made a fresh and lively appearance as a Shinhan Life model and captivated so many minds, is a good example.


This AI model Rosie has stirred a sensation among many followers.

However, machine learning which is a subset of AI does not apply to just such ‘professional’ examples but also so many different aspects of our mundane lives. This segment will list different use cases of machine learning that would be interesting and comprehensible to everyone.


1. Netflix: Recommendation algorithms, we are already aware of, but the artwork personalization?

Due to COVID-19 and the vastly increased stay-at-home time, the OTT (Over The Top) market has been expanding almost explosively. Just like how ‘Googling’ has become a widely used term, ‘Netflix and Chill’ has also become an expression everyone understands. Korea has produced several hits on Netflix too, such as ‘Squid Game’ and ‘Kingdom’. Machine learning technology is used in Netflix to offer consumers series and shows according to their taste.

Netflix makes lists of shows, keeping track of the user’s picks

Netflix displays the recommendations for the users based on the machine learning algorithms by reviewing the user’s preferred shows and grouping their information by similarities. Then they sort other shows that are similar to the saved data. A personalized list of Netflix shows is put together this way. However, even many of those who aren’t familiar with machine learning are aware of this mechanism.

But then, had you acknowledged that there was also artwork personalization on Netflix? The artwork displayed as a sort of thumbnail varies according to the user’s taste. Machine learning comes in here as well.

The recommended artwork differs based on the viewing records of the user.

If the user is into romance, the recommended artwork appears with couples in them. If the user prefers a bright, vibrant, and comic genre, the artwork will contain a lively-looking atmosphere in it. What the viewer has watched so far forms a certain list and then the pattern is reflected in the artwork display.

2. Pepsico’s Cheetos: Machine learning to maintain the crunchiness of the snack during the manufacturing process

Cheetos is a snack widely loved all over the world

A leading food-making corporation Pepsico applies reinforcement learning, a method of machine learning, for the maintenance of the Cheetos’s crunchiness in the best condition. With reinforcement learning, the size, volume, distribution of additives are kept even-textured and uniform. This task takes a few days for a human to handle, while the AI solution used here takes around 30 seconds. Apart from this, machine learning has also been incorporated to devise a strategy focusing on expanding the market through demand forecasting, such as referring to demographic data to additionally stock popular products in specific stores.


Machine learning was used to implement the ‘perfect crunchiness’ of Cheetos


3. Wikipedia’s 2-months-worth forecast of web traffic: Machine learning was used to draw out the 65 days-worth web traffic in advance

Wikipedia is a search engine to pass by essentially for different uses, from simple searching to even schoolwork (although you might want to leave it out of the citation source). The cumulated amount of data by the users so far would be incredulously influential once used for other purposes. A machine-learning company promoted a forecast project of predicting two months’ web traffic with a machine learning method, time series analysis. Two years’ worth of search records in chronological order were put into the machine, which then produced future predictions on the estimated amount of use by different sites.


Machine learning informs how much traffic each website would receive in the future


4. Behavioral Biometrics Authentication: Machine learning analyzes device usage patterns

Have you ever heard of behavioral biometrics authentication? Everyone has slight variations in typing speed, mouse-clicking force, swiping speed, and direction. Behavioral biometrics authentication is utilizing this diversity of user action for identity verification. Such authentication reinforces and enhances the quality of security, beyond mere password-matching programs. It compares the saved behavioral patterns of the user with the detected user pattern and refuses access if there is a significant nonconformity. Machine learning is used in this elaborate distinction process.

Everyone holds unique behavioral patterns, and machines remember that. 

Picture source:


5. Fox Sports Australia: Would you like to be transported 5 minutes later on in the cricket match?

Are you aware of the sport cricket? Fox Sports Australia, a sports channel group, has ambitiously started a project that predicted whether the batter would be put out or not in 5 minutes from the ongoing game. With a year’s worth of cricket game information, the machine learned approximately 83 factors involved with the batter hitting the ball. Predicting the probability of the batter being put out 5 minutes before the real situation through machine learning, all based on past experiences! Isn’t it then pretty safe to say that predicting anything and everything but lottery numbers is possible?

In cricket matches, the batter is put out once the ball knocks the wicket off.


6. Medical service: Detecting any danger with machine learning in advance

Every fatal disease requires attentive and careful concern. In the medical field, defining machine learning to be saving people’s lives would not be an exaggeration. Machine learning can be used to catch any strange symptoms or patterns of the body with diabetes or heart diseases. The sensor keeps monitoring the conditions and sends an alert to the medical staff if there is an irregular pattern or an intensified symptom. If a particular sign is constantly observed, the machine sends a signal judging that it is likely for the patient to suffer the upcoming worse cases. Once such technology becomes more elaborate, failure to detect danger by the medical staff will decrease.



7. Sunday Toz, predicting who would make in-app purchases in Anipang

The biggest profit-making channels for game companies are in-game purchases and advertisements, which is why it’s important to distinguish those who make in-app payments from those who don’t. Machine learning is used here to calculate the probability of conversion to the purchase. SundayToz, a company that rose to the surface of fame with ‘Anipang’, the game that has hit jackpot, carried out a project to predict customers making payments through collecting and training on 37 billion cases of data gathered over 3 years. A segmentation of advertisement exposure was practiced, where certain advertisements were aimed towards customers likely to pay, whereas ad exposure was modified among those likely to enjoy the free features. The goal was to excellently perform as an advertising platform without changing the existing sales structure.

Try segmenting the users into those who would pay and who would not! 


At this stage, meticulous work was put into the process to distinguish between significant and insignificant variables among the vast data. Here, human judgment acts as an important role player in machine learning. Because there is a lot of information and so uploading it haphazardly could lead to a low-quality answer, it is necessary to extract meaningful data and remove unnecessary data because it may impair the accuracy.

◆ Selecting which data to ‘ignore’ is an important task. However, if there are hundreds of columns of data to be input, it will take a lot of energy to sort out the necessary information one by one. DAVinCI LABS provides the ‘Auto Selection’ function among the types of input data to select important and can-be-overlooked variables. In addition, it calculates the correlation between input information columns (types and variables) numerically or displays them visually. It is to present an organized flow from the point of view of data science so that non-experts can sufficiently create a predictive model.


DAVinCI LABS gives recommendations on the ‘input field(variable)’.


You can examine the correlations between the input variables with the confusion matrix.


8. Financial Industry: Preventing fraud, detecting anomalies, and CSS (Credit Scoring System)

One of the main uses of machine learning in finance is fraud detection, which prevents the fraudulent acquisition of other people's money. The data used to detect fraud is the customer's personal information. As it deals with sensitive information, the existing fraud detection and prevention system takes up a lot of time and has technical limitations in only catching clearly fraudulent cases. The introduction of machine learning is effective in accelerating business processes by quickly and accurately detecting fraud.

Fraud can be detected based on the client’s transaction data!

Let's take an example of detecting credit card fraud. The machine gets trained on the data by the actual client: transaction information, IP address, transaction date, amount, product type, and so on. If someone is trying to proceed with fraudulent transaction, it surely will be difficult to perfectly mimic the client’s usual transaction patterns recorded up to date. Obviously, there will be cases with some discordance compared to the previous, genuine records. It could be the IP address, it could be the transaction amount and other consumption habits. When such unusual behavior is detected, the machine determines that it is fraudulent. Through machine learning, the machine can inform you in advance of parts that humans may overlook.

Details on how machine learning can be used to detect fraud are presented in a previous post on Pitney Bowes. Please refer to the Lending Club post for credit rating-related information.

<Usecase of fraud detection: Pitney Bowes>

< CSS use case of loan repayment prediction via client data: Lending Club>

DAVinCI LABS, an AI AutoML Solution

DAVinCI LABS is an automated solution that creates predictive models through supervised learning and time series analysis in machine learning.

As an artificial intelligence machine learning automated solution, the field that is currently being widely used is the financial industry where customer data is relatively well-structured. Due to the current situation of global financial companies, where it is difficult to recruit and nurture technical talent, DAVinCI LABS can apply it to the field based on data that only a few could process and utilize. There are not many companies that provide automatic machine learning solutions in Korea, but Ailys makes data analysis and predictive model implementation possible in various fields through DAVinCI LABS. The biggest point is that with a few clicks, you can create high-performance machine learning predictive models that have been previously reserved for only the experts.


DAVinCI LABS’  Use Case in the Financial Industry

Gradually you will start to realize that AI is no longer a merely vague and distant subject only to be accessed through the media. In particular, the non-face-to-face culture is spreading due to the prolonged COVID-19, and communication methods such as AI interviews are becoming more common. Through a collection of simple examples, we explored how machine learning as a part of AI affects our daily lives.

In the upcoming post, we will look at an example of the financial use case of DAVinCI LABS and how it could benefit the industry. Keep up with our updates!