In today's world, most advanced organizations use machine learning technologies in their business processes. Machine learning, a subset of artificial intelligence (AI), enables algorithm-based software to perform various tasks such as optimizing resources, tracking revenue streams and managing digital tasks. Initially, this requires an experienced programmer to take on tasks such as creating, training and optimizing machine learning models. However, automatic machine learning (AutoML) technology can greatly simplify this process. AutoML software can automatically select and train the most effective model without the need for an expert.

So what exactly does automatic machine learning mean and what benefits does it offer companies? Let's get to the bottom of these questions together

What is Automatic Machine Learning (AutoML)?

Automated machine learning (AutoML) refers to technologies that use software platforms to automate the selection, implementation and data science processes of machine learning (ML) models. Since the algorithms are automatically generated by the software, no expert is needed to programme them. As it is much faster than conventional machine learning, data science experts can save a lot of time. It can also deliver more accurate and reliable results compared to manually coded processes.

The process of automatic machine learning consists of five basic phases. Let us take a closer look at them.

• Pre-processing of the data: Basic data preparation processes such as adding missing data, organizing and scaling are automated.
• Model selection: AutoML software automatically selects the most appropriate model among different machine learning algorithms based on various factors.
• Hyperparameter tuning: Various hyperparameters that determine the performance of the model during training need to be optimized. Running sets of hyperparameters helps in creating the structure, function and performance of the models.
• Model optimization: Various techniques are used to improve the performance of the model and optimize the training parameters.
• Evaluation: The performance of the trained model is evaluated using various metrics such as cross-validation.

What advantages does it offer companies?

AutoML software can manage many complex business processes independently. For example, it can optimize back-office processes such as inventory management, document processing and cost tracking. It can monitor supply chain data in real time. It also eliminates the need for data scientists to keep experimenting to find the best performing model in a data set.

With AutoML, companies can save a lot of time and budget by automatically processing data sets. The software allows models to be selected and tested multiple times within a short period of time. This allows data scientists to focus on other tasks instead of selecting, training and optimizing models.

Another advantage of automatic machine learning is reliability. Since analyzes that are traditionally performed manually by data scientists are carried out by the system, the margin of error is reduced. This leads to high accuracy rates.

At SOCAR Türkiye we closely follow the developments in the field of machine learning technologies. We focus on continuously improving our digital capabilities in this area, as well as in many other areas of technology. In this context, we have developed SmartXSensor, an AutoML tool, within the R&D and Innovation department of SOCAR Türkiye.

SmartXSensor not only has features such as model creation and data pre-processing, but also has better performance than other AutoML software due to its variable selection algorithm. It greatly simplifies model development in processes that need to be monitored with a large number of sensors by automatically selecting the most suitable variables for machine learning models with high predictive power.

“Our SmartXSensor with its features such as model building and data pre-processing as well as its variable selection algorithm shows superior performance compared to other AutoML software.”

As part of the DCU project to optimize coke output, we managed to analyze extensive data from more than 500 different sensors over three years to model the processes of converting the heaviest part of the crude oil into two main product groups within the DCU plant. We used this historical data to train 17 state-of-the-art machine learning models and an optimizer. We customized the AI-powered data-driven decision-making process for both the business level and engineers to make key strategic decisions. The advanced analytical models enabled the engineers to optimize the efficiency of the DCU units and ensure proactive control under ever-changing refinery conditions. As a result, our project reduced coke output by 2%, increased capacity utilization by 1% and achieved 100% compliance with LPG quality targets.

The optimization model created by our digital transformation team using advanced analytics and machine learning techniques tracks the production process, collects real-time data, processes it, runs simulations and offers suggestions for improvement to the employees involved in the process. This reduces waiting times while implementing data-driven decision-making mechanisms in real time.

As a pioneer in the industry and a forward-looking energy company, we will continue to follow the latest developments in research and development and innovation and develop innovative solutions tailored to our areas of application. Because we work today with the responsibility to leave a lasting impact for tomorrow.