The 4th Conference on Machine Learning and Systems (Virtual-only Conference) was held online from April 5 to April 9, 2021. Since its inception in 2018, the conference has been classified as a top meeting. Digital China, along with the leaders in ML (machine learning) and System (computer systems), showed up to demonstrate its multi-industry applications of AI model and algorithm, and systematically explain the way to explore business principles and create values with mathematical models and algorithms.
Global data are booming in size. According to statistics and forecasts by Statista, data produced globally in 2020 were 47ZB, while the figure will reach 2142ZB by 2035. Consequently, the size of global data is about to explode, and the utilization is growing by leaps and bounds. Data have become the core assets of enterprises as the basic means of production. Effective management and use of data assets are crucial to the digital transformation of enterprises.
Supported by data, business management knowledge, and data analysis skills, DC makes use of mathematical models and algorithms to explore business principles, backs up management decisions, creates value for leading customers in different stages of digital transformation, monetizes data assets and empowers digital transformation. The practices and applications of AI model and algorithm in automotive, hotel, retail, finance, aviation and other industries were also demonstrated at MLSys.
Automotive：Aftermarket supply chain optimization through full life-cycle management of components
Data collection and KPI monitoring are conducted during supplier production, logistics, inventory turnover and other processes in the supply chain of auto parts. A total of 140,000 parts in circulation of an automobile brand are predicted according to four periods, namely launch period, growth period, stable period and decline period. The overall precision of parts climbs by 3% or so, and that of parts in the decline period (breakpoint parts) and stable period (mature parts) rises by more than 5%. the overall prediction of 140,000 circulating parts of an automobile brand is made. The yield ratio maintains at 95%.
Hotel：Foot traffic-based revenue management and dynamic pricing
DC has formed a thematic data mart oriented at revenue management, comprising the data from the systems governing rooms, customers and finance, as well as those from weathers, holidays, events, and marketing policies. The room rate is determined by foot traffics which are predicted via multiple regression model, ARIMAX, ESM, LSTM, etc. The error rate of foot traffic prediction across 93 branches of a hotel brand is about 1.4%, and the overall revenue increased by 12% when application windows are predicted.
Retail：Mechanism of refined oil pricing through price testing
DC has cleaned and integrated the data from the fuel card system, financial system, supplier system, HOS system of refined oil retailers. Groups are divided into as per the daily/annual purchase, gas station, price policy and market campaign. The sales volume and gross profits corresponding to the results will be calculated to determine the principal factors that affect revenue, gross profit, and pricing during the price testing period. An optimal pricing strategy for purpose of maximum gross profits will be figured out based on the analysis of pricing records, JV sales volume and gross profit, purchase price, volume change, competitor's price and other factors.
Finance：An anti-fraud engine boosted by a rule model and SNA
A rule model is built by DC based on the information of loan applicants, so as to mitigate potential risks. The social relations of applicants and people on the blacklist or grey list will be analyzed by SNA and scored by correlation coefficients. With the aid of the rule model and SNA, the decision-making engine will be informed of the results to judge whether an approval or a manual audit by a risk manager is needed, so that financial institutions can avoid financial risks.
Aviation：Meal solution optimization with the cost simulation system
DC has thoroughly analyzed the flight variable cost and its influencing factors and marginal cost. Exampled by the meal campaign launched by an airline company in 2019, more than 3 million pieces of flight information and 30 million pieces of meal data stored in CMS over the past five years were processed by DC. Characteristics of flights and passengers are analyzed from the perspective of reducing food waste and cost, while probabilities are calculated from the perspective of space, departure time, flight time, route, etc. The meal strategy and cost monitoring system made by the cost simulation system save about 10% of the meal cost for short-haul flights.
It is of great importance for companies to seize the opportunities of the digital economy by using models with higher speed and accuracy to identify profitable opportunities and avoid unknown risks. DC will optimize data models and algorithms based on business scenarios, implement multi-industry digital solutions, tap the value of digital assets and monetize them in a secure and compliant manner. 《/