金年会|金年会官方网站

AutoML

Our AutoML platform is the world's leading end-to-end AutoML platform designed for users of various skill levels who are interested in using machine learning to solve business problems effectively. It is extremely easy to use, high quality, high speed, and provides flexible modeling modes: automatic, advanced, and manual.

The platform's end-to-end high-quality automated model building feature helps users to focus solely on solving business problems rather than on machine learning problems. It helps business analysts who know data and business logic well to become efficient “data scientists” quickly and unleash the power of their business domain expertise without having to go through lengthy machine learning training programs. It can also effectively assists good data scientists to speed up the model building process and develop better quality models that fully leverage the value of the data.

The AutoML platform platform consists of two main modules: model training and model operation, and each module contains numerous functional features.  

The following figure intuitively illustrates Our AutoML platform’s basic working logic and internal functional modules.

Main function modules of the AutoML platform platform

1.1        Extreme ease of use

Our AutoML platform’s a fully automated, end-to-end machine learning dev & op platform that offers a one-stop service from data cleaning to model building and deployment with only a few clicks of buttons. Whether the users are experienced machine learning experts or ordinary business data analysts, they only need to follow the guided GUI and quickly complete all the tasks to build high-quality ML models. With that, Our AutoML platform greatly reduced the formidable barrier of machine learning, and machine learning is no longer the privilege of a very few well-trained experts.

Intelligent user interface of the AutoML platform platform

1.2       Superior and reliable model quality

Thanks to Our AutoML platform's world-leading end-to-end automation and global optimization technology, Our AutoML platform has the best model exploration capabilities that are far beyond human. In most of the real-world use cases, Our AutoML platform built average 11% better models on average than the current manual process. Moreover, the platform can also continue to improve the model quality automatically, even after the models are in operation. 

1.3       High model development speed

With the assistance of Our AutoML platform platform, the time needed to train a model can drastically be reduced from several weeks to hours. It helps shorten the overall project development cycle, reduce the overall project cost, increase the chance of success, and quickly meet the business opportunity windows.

Modeling Efficiency of Our AutoML platform Platform

APPLICATION SCENARIOS

Bank

Fraud identification

Artificial intelligence anti-fraud applications can effectively block all kinds of fraud risks, including transaction fraud (counterfeit card/stolen accounts, unauthorized use and bank account transfer, etc.), illegal cashing, card recognition for anti-money laundering, application fraud, tax rebate fraud, insurance fraud, and fraud gang, and many other scenarios to help banks identify malicious users and behaviors. It also can solve fraud threats in payment, loan, and risk control, and consequently minimize corporate losses.

Prediction of loan repayment ability

Through the lender's historical data, such as a lender's annual income, collateral, historical loan information, and related person income, the machine learning model can predict the lender's repayment ability.

Intelligent investment

According to the risk tolerance level, income target and preferences provided by individual investors, a series of intelligent algorithms combined with optimization and theoretical models are applied according to the goals, age, income, and current asset status of the user, smart match in assets and financial instruments, to make real-time adjustments based on market changes.

Credit evaluation

Based on the user's basic information, historical consumption, and historical repayment, the platform can transform the original data into feature engineering, and establish a scorecard model to evaluate the credit risk of the user.

Quantitative transactions

Through thousands of stocks’ historical price data, the artificial intelligence model can be used to distinguish stock portfolios that might rise and those that might fall in the future. According to the prediction results of intelligent algorithms, a company can cyclically trade hedge stock portfolios, hedge risk different stocks and seek arbitrage opportunities.

Smart service

By predicting the customers need or status, intelligent customer service can conduct customized business consultation and provide better customer experience.

Insurance

Insurance Pricing

Featured data is selected based on driver information, driving behavior, existing insurance, vehicle information, and historical claim records. The platform can build machine learning to improve model differentiation, maximize profitability, and provide support to customers.

Smart underwriting

With a risk model constructed through artificial intelligence, the risk level of the case and the corresponding payment method can be determined in real time after the user applies for a claim,

Telemarketing / Precision marketing

Based on the user's basic information, vehicle information, and historical records of success telephone sales, variables that have an important influence on the telephone sales success rate are selected to determine user profile. With the predictive model built based on the user profile, the success rate of telemarketing will improve.

Insurance innovation

Based on user scenarios, insurance companies can make innovations such as P2P insurance, real-time insurance service based on wearable devices, smart homes, and other devices, to help companies innovate insurance pricing and predict insurance revenue.

Driving behavior score

Based on vehicle maintenance data, traffic violation data, weather condition data, and driving data, semi-supervised or unsupervised machine learning methods are used to develop a driving behavior scoring model to provide more accurate pricing factors for vehicle insurance.

Healthcare

Predictive analysis of diagnosis and treatment plan

According to the patient's situation, predicting the cost, efficacy, and risk of the treatment plan will help doctors choose the best program, improve the treatment outcomes, avoid risks, and use medical resources more effectively.

Personalized precision medicine

Based on the patient's basic information, historical physical measurement data, historical medication status, and other information, artificial intelligence algorithm will generate a personalized health management program for the patient automatically according to the patient's condition,

Disease prediction

Based on the genomic data of users, machine learning can be used to achieve high accuracy in identifying single nucleotide polymorphism variation and insertion/deletion (SNP and INDEL), copy number variation (CNV), and assessing the severity of genetic variation.

Clinical trial matching

Based on patient personal medical history and genetic information data and combined with clinical trial recruitment data, artificial intelligence algorithms are used for rapid matching to accelerate the registration of clinical trials for cancer treatment.

Chronic disease management

Through data acquisition devices, such as wearable devices, artificial intelligence is used to make decisions and provide users with personalized chronic disease management recommendations.

Mental health management

Analyze the relationship between user behavior and psychology by collecting user data and using machine learning modeling,

Smart consultation

Artificial intelligence can help doctors respond to patients online by establishing a disease knowledge base and historical interview records.

New drug development

Extract the knowledge that can promote drug development from massive information stores and make predictions to accelerate the drug development process.

Generic drug development

Using artificial intelligence can help modify designs based on new drugs abroad to find similar chemical structures that are not protected by patents.

Game

Game retention data mining

Use the pre-game buried point data to find features that maximize the difference between the remaining players and the lost players and improve the user retention rate.

Paying game user prediction

Find unique characteristics of paying players and non-paying players from the player's game behavior data and predict paid users by machine learning, improving the player's payment rate and payment depth pertinently.

Game chat advertising filtering

Using various machine learning algorithms, the company can train a proprietary advertising recognition model for each game product, accurately identify the advertising speech of the user's chat channel, and identify the advertisement and the publisher in real time.

E-commerce / New retail

Recommendation system

The company can customize the recommendation system for customers based on a variety of machine learning recommendation mechanisms such as product similarity, association rules, collaborative filtering recommendations, etc.

User profiling

The company can improve ROI by tagging the basic information of users and recommending product in a personalized manner according to the granularity of their behavior.

Prediction of explosives

Based on the commodity historical data set, machine learning can create a prediction model for explosives, which improves the rationality of business planning and maximizes GMV.

Customer churn prediction

Based on churned customer data and retained customer data, using a variety of machine learning algorithms, companies can build a customer churn prediction model, forecast current customers, and retain outgoing customers ahead of time.

Potential customer identification

Based on a variety of data sources such as ad delivery data, clues, CRM data, etc., the company can analyze the correlation degree of potential customers and predict the probability of customer turnover.

Precision marketing

Using its advanced AI technology, the company can automatically capture and analyze characteristics of users’ behavior, and therefore quickly implement personalized marketing approaches on thousands of people.

Smart customer service

Based on the user's basic information, product purchase information, and asked questions, building a Q&A knowledge base benefits smart customer service in improving customer satisfaction.

Energy

Power generation prediction

Based on a large amount of accumulated historical data, using a machine learning platform to model and establish a predictive model helps establish a deep, wide-range, multi-time scale and renewable energy prediction system.

Demand prediction

Different methods are applied for different types of load prediction, and load prediction models are constructed in combination with the development and land use planning of the planning area. Season studies typical daily load curve of each energy product and curve fitting generates the typical daily load curve of the planning area. The company can use machine learning to predict future energy daily loads and update the model in real time with information collected from different users.

Resource optimization

Analyze and classify the characteristics of meteorological elements corresponding to renewable energy. Comprehensively consider the terrain slope, surface cover, land use planning, etc., to calculate the available area. Predict the annual power generation according to the characteristic parameters of the energy, such as the wake effect of the wind turbine and the installation method of the photovoltaic to realize the optimal allocation of resources.

Electricity theft detection

Build machine learning model based on user's electricity consumption data, characteristics such as voltage, current, daily electricity consumption, power consumption change, and line loss variation, to identify the tampering behavior by abnormal value detection.

Telecommunications

Customer churn prediction

Based on user behavior, relationship network, complaint mobile search, and payment, defining leaving users’ characteristics and using multiple machine learning models to predict whether users will leave the network, thus making customer retention recovered in advance.

Recommendation system

A variety of recommendation mechanisms based on product similarity, association rules, collaborative filtering recommendations, etc. are used to customize the recommendation system for customers.

User profiling

The company can improve ROI by tagging the basic information of users and recommending products in a personalized manner according to the granularity of the user's behavior.

Potential customer identification

Based on a variety of data sources such as ad delivery data, clues, CRM data, etc., the company can analyze the correlation degree of potential customers and predict the probability of customer turnover.

Precision marketing

Using its advanced AI technology, the company can automatically capture and analyze characteristics of users’ behavior, and therefore quickly implement personalized marketing approaches on thousands of people.

Smart customer service

Based on the user's basic information, product purchase information, and asked questions, building a Q&A knowledge base benefits smart customer service in improving customer satisfaction.

Telecommunication fraud prediction

The company can build outlier detection machine learning models on user communication, complaint, and relationship data, etc., to identify telecommunication fraud.

Maintenance of telecommunication equipment prediction

The company can build a classification or a regression model based on telecommunication equipment operating data, sensor data, runtime, operational status, etc., to predict whether the equipment will fail or predict when the next failure occurs.

Manufacturing / Automobile

Predictive Maintenance / Product Life Quantification

Use machine learning modeling of sensor operating data on key parts of each device, accurately predicting the life cycle of different types of equipment. Find outbreak pattern to make early warning on equipment failures, identify the operation and maintenance cycle of each equipment, and predict spare parts procurement.

Defect Detection

Through deep learning algorithms, the company can achieve unmanned detection by having deep learning model to identify any defect in products.

Quality Control

Based on attribute data in the manufacturing process, machine learning algorithms are used to predict product compliance and select key process parameters that affect product quality.

Customer monitoring

With the company’s detailed collection of information on manufacturing companies, machine learning can be used to predict their ability to pay and provide real-time warnings.

Autopilot

A variety of supervised learning algorithms and unsupervised learning algorithms are used to perform target detection, target classification, target localization, and motion prediction, which are applied to the development of autopilot systems.

Quality inspection time prediction

With historical data of on-road tests, machine learning algorithms can predict the time required for testing new cars with different models and different configurations, and help rationalize the designs of new cars.