If someone inquires what solution DAVinCI LABS is, I would refer to it being an ‘Auto Machine Learning Solution’. Machine learning finds patterns through data training then builds a model to predict the future, whereas auto machine learning helps beginners to be able to get through the whole process.
As mentioned in the previous content a few times, traditional machine learning techniques require advanced skills and experience all the way from data feature engineering through model development to its management. That’s why, in order to utilize machine learning, most companies had to ask for exterior management or hire an expert and pay for the consequential resource.
However, machine learning models require constant update and management including new data input and training, considering that they were initially designed to make a precise prediction of the future. Also, if there are more than one model needed from each business department, it’s prone to involve too much time and money to carry on with the general, traditional method of hiring exterior skillset.
AutoML solution was created on top of this status. Not just data scientists, but those working in real business fields can develop and manage the model handily and make real-time future predictions.
There are four conditions to be fulfilled in order to be called an AutoML solution.
1. Interface should be simple and easy to understand even for new users.
2. There should be no separate need for coding.
3. Even the most advanced features should be open to auto processing.
4. The whole process, including the completion of model development, should be simple.
Nowadays we could frequently come across articles about companies that have developed their own automated machine learning solutions. However, in reality, it’s not easy to find solutions sufficing all four of the mentioned conditions. Expert knowledge and skillset became necessary for advanced operation, as ‘Auto’ features were applied to only a portion of the product. In other cases, coding skills were sought after.
However, DAVinCI LABS fulfill all four of the conditions, proving itself to be a real AutoML solution. You can confirm that coding is unnecessary for model development through the video below.
In this case what types of modules have DAVinCI LABS automated? We’ll first take a look at the predictive modeling module.
1. Predictive Modeling Module
DAVInCI LABS’ Predictive Modeling module consists of four segments.
Auto Modeling: Once a prediction target is set, algorithm helps learn data patterns then to create a prediction model (e.g., Makes a prediction on whether Client A is likely to pay back the loan)
Rule Generator: It analyzes existing data to make clusters of high or low target values (e.g., Analyzes characteristics of a cluster that contains clients with high loan repayment rates)
Time Series Analysis: If the target to predict changes over time, it creates models according to time flow (e.g., Predicts product demands on a specific date)
Auto Clustering: Even through absence of a specific target, existing data gets analyzed and then grouped into clusters with distinct differences. (e.g., Points out customers who are most likely to commit fraud or crime)
The module with these four segments could each be matched to either supervised learning (that has a clear target) or unsupervised learning (that doesn’t state a target). You can read the piece below to learn more about supervised learning, unsupervised learning, and clustering.
DAVinCI LABS is the one and only AutoML solution that holds all features: supervised learning, unsupervised learning, time series analysis and clustering. Machine learning, generally known, tends to contain only supervised learning or modeling without clustering. DAVInCI LABS is a solution that well incorporates Ailys’ values: pursuit of not just automation but a practical and universal AI technology.
Not only does DAVinCI LABS have a development module of prediction models, but it also carries an optimization module aiming effective decision-making in business. We call this a Decision Optimization Module.
2. Decision Optimization Module
DAVInCI LABS’ Decision Optimization Module consists of three segments.
Simulator Optimization: When applying new data onto predictive modeling, it optimizes the variable value so that it could reach the desired target value. (e.g., To increase the loan repayment rate of Client A, current credit line should be lowered by 30%)
Rule Optimization: You can see how the target value of the cluster would be modified when there are adjustments made to the size and variables of the cluster. The cluster is a result of the created rule’s classification. This segment also offers the optimal variables required to satisfy the target average value of the cluster. (e.g., If a new variable is added to the client group with a high loan repayment rate, Rule Optimization gives information on the effects the repayment rate would have on the repayment rate, and which variable must be added to increase the target value in average)
Cluster Optimization: It either fragmentizes automatically created clusters or inputs feedback on some samples to optimize the clusters for business. (e.g., Reflects characteristics of clients likely to commit fraud or crimes to build a clearer cluster)
The reason for carrying such optimization modules separately is clear: a machine learning prediction model cannot cover the business strategy itself; it can only provide future predictions.
For example, let’s assume a situation where an insurance issuance evaluation model predicts the probability of a client’s insurance invalidation or cancellation. What business strategies would we be able to draw from the prediction model? The simplest one would be discouraging the client with high probability rates from signing the insurance contract. However, there could also be ways of recommending a different product or modifying certain conditions within the insurance contract to lower the possibility of client churn. In other words, if the consequential customer behavior to a certain action is predicted in advance, we could come up with the optimal action to take for this. This is precisely a business strategy incorporating a machine learning model.
To wrap everything up, DAVinCI LABS could be defined not just as a mere AutoML solution but also something that can optimize decisions. This is referred to as the Next AutoML : Adaptive Intelligence. It implies that the solution is capable of adapting to various circumstances and simultaneously draw out the optimal strategy. Artificial intelligence imitates human intelligence, machine learning imitates within that learning and predicting abilities, and adaptive intelligence is a skillset that can align the drawn information with business/market status to generate the optimal answer.
By now it’s likely there would be some of you who have grown curious about how actually each module gets executed. We’ll come back soon with our demonstration videos!
