Machine Learning applications

Detailed below are some of the recent machine learning applications.

Agriculture and Farming

About 8000 B.C. ago, at the dawn of agriculture, the total number of living human beings on Earth was about 5 million. According to the United Nations, on July 1, 2015, the world’s population was 7.349 billion and is projected to reach 10 billion people in the year 2056. Studies estimate that global agricultural production needs to increase by 60 -110 %  by 2050 to meet the growing demand. Agriculture depends on many volatile factors such as geography, climate, biology, economics, and politics. Furthermore, land use involves multiple interactions making it a complex system with environmental constraints.

Precision agriculture uses machine learning to feed people by providing a set of tools to gather information and allows farmers to make proper and place-specific decisions about agricultural inputs, processing, and production.

Scrutinizing the phenology of plants improves the timing of harvest, yield prediction, pest control, and disaster warning. Supervised and unsupervised learning methods use weed discovery based on shape descriptors and early detection of plant diseases based on spectral structures. Supervised self-organizing maps to handle existing information from different soil to analyze multi-layer soil facts to figure crop growth and yield.

The recent use of machine learning in geography is Cellular Automaton (CA) to study land-use change. Based on the states of neighboring cells, CA collects cells on a grid of stated shape that changes according to sets of rules and through separate time steps. For example, ThingWorx uses machine learning to automate advanced and predictive analytics for irrigation systems and a network of sensors, to avoid dry soil, the growth of fungus, or avoid water regulations in California. The technology learns from data, discovers patterns, builds validated predictive models, and sends information to any application.

In the Netherlands, the Financial Economic Simulation model affects the agriculture sector. However, during a review, it was found that the model is static and does not take into account structural changes. Models using Multiple Linear Regression and Neural Networks built and predicted the farm sizes change effectively four-year in the future.

Machine learning methods are applied in the aquaculture domain to predict shellfish farm closure, predict algae bloom, set up a new farm, and produce habitat maps from seafloor images.

Other uses of Machine Learning applications in the Primary Sector such as :

Mining, Quarry, and Archaeology

The mining industry uses machine learning for smarter explorations. The data gathered help in identifying the minerals, the chemical composition, and detecting even deeper deposits of minerals. The modeling of mineral deposits and economic impact during the early stages helps mining companies save time and money. It provides safe working conditions for miners and checks the real-time flow of the rocks during blasting operations. For instance, in Iran at Sungun copper mine, statistical analysis techniques predict fly rock distance to help in minimizing the related health hazards and other undesirable environmental impacts such as air overpressure and ground vibrations.

Fossil hunting uses neural networks to learn the pattern and make predictions about the data. An example is the use of networks in the Great Divide Basin to distinguish fossil sites through satellite images. The model recognizes correctly 79% of the pixels that were known to represent fossil locations and 99% of the pixels flagged by the network held fossils.

Four stages classify the industrial revolution. The first stage was the mechanization of production using water and steam power followed by mass production with the help of electricity. The next stage was a digital revolution with the use of electronics and information technology to automate further production. Industry 4.0 or ‘smart factory’ is the subsequent stage facilitated by machine learning applications in the secondary sector to produce systems that monitor physical processes and communicate with each other and humans in real-time.

Automobile Production

The data about the driving environment obtained through many sources such as onboard cameras, global positioning systems, or engine combination failures help intelligent systems in varied fields of the automotive industry. Examples of these intelligent vehicles technologies include the Aston Martin DB9 engine control system, the use of the Direct Labour Management System at Ford Motor Company to enhance assembly process planning, standardize labor time required and meet consistency of vehicle process manufacturing. Ford also uses data mining techniques to find occasions to cut materials from a supplier.  Hence, these systems help in improving the operative productivity of the assembly process and reduce manufacturing costs.

Jaguar Land Rover is using the latest machine learning and artificial intelligence techniques to propose a “self-learning car” with the ability to learn preferences and driving styles and offer a broad area of services to the driver. Jaguar’s smart assistant will be able to plan the car navigation according to calendar schedule, adjust car settings and adapt car temperature after a certain activity by the driver. Google’s driverless cars and Telsa Motors use the software’s ability to recognize objects in different road environments to facilitate the driver to hand over control under specific conditions and help in making smart driving judgments. Thus, the system helps in reducing collisions without eliminating the fun of driving.

Textile Production

Some of the uses of machine learning applications in the secondary sector are classified as:

Construction

Construction, engineering, and management use artificial neural networks through predicting risk and optimization, classification and choice of resources. Some examples are training neural networks to study basic soil properties from a dataset of different cohesive soils from various locations and solving engineering design and image recognition by pattern recognition.

Food Processing Industry

An example of machine learning in the food processing industry is the broiler industry in Iran. Iran is one of the top 10 producers of chicken meat in the world and uses machine learning for making models of important environmental trends such as prices of maize, soybean to predict the cost of chicken meat and product supply chain. The trade ministry regulates the market in case of deviations to give food and save jobs.

Shipbuilding

Machine learning techniques such as non-linear optimization, hierarchical cluster analysis, and principal part analysis estimate costs, for example, The North Atlantic Treaty Organisation (NATO) estimated the cost of amphibious assault ships in the Netherland.

Breweries

Due to the process involving the use of living organisms and biological raw materials, it is difficult to set up a control in the drink industry. Daily measurements assist in monitoring the process and keep it to a desired course of fermentation.  Neural networks provide a mechanism to help monitor and control fermentation. Also, reliable predictors lead to fewer measurements and early warnings to make corrective actions. The relationship between flavors and the sensory perception model predicts the flavor of certain types of beer.

Energy Utilities

Machine learning can help demand-side management by reducing computation time and predicting energy consumption. Networks identify faults through pattern recognition and efficient data analysis. It also helps in diagnosing performance by identifying issues in electrical equipment and address fluctuation in power generation. An example using machine learning is a California-based firm called App Orchard which performs analysis using simple natural language queries to manage transmission and workforce.

MetalWorking

Artificial neural networks help manufacturability exploration, process planning, and die design. Finite element methods find out the non-linear relationship between design and input parameters. Analysis and modeling of the metal-cutting process determine ideal response values and, therefore, increasing the product’s quality and life of the tools being used.

Transportation

Machine learning techniques, for instance, probabilistic graphic models help in obtaining patterns and models from data by describing multivariate continuous densities. Real-time data model traffic patterns to support inferences and support large-scale traffic simulations.

An example is the use of algorithms to discover real-time hot spots for congestion in Sydney CDB, Australia. Also, it predicts near future traffic jams due to an incident on the road. As efficient performance and management process depends on estimating the total duration from response activation to clearance time, the ideal route for emergency vehicles avoids congestion to reach a destination. Traffic watch microscopic simulation saves 40 percent of travel time during peak hours or if applied to the whole motorway system in Sydney there is a potential for savings of more than Australian $ 500 million per annum.

Maritime transport is the backbone of international trade as three-quarters of global trade by volume is carried by sea. Due to an enormous increase in the data being collected by the electronic tracking system, automatic analysis improves situational awareness in the maritime domain. Lack of knowledge is made up by learning from data. Also, machine learning helps in improving the detection of anomalies.  A perceived situation is normal if it is predictable and the rest is anomalies.

Entertainment

Every aspect of the entertainment industry uses machine learning. It is used in online advertisements, measurement of audience accurately, media pricing, product distribution, and content delivery. Social graphs help in analytics and predicting user behavior with relevant messages yielding a three times increase in users’ engagement. Also, automated categorization generates classifications for digital media say news articles in Reddit. Sites use recommendations to users such as Pinterest, which uses multi-class image categorization to suggest boards.

Likewise, market research analysis improves the application’s notifications sent to the users. In video games, designed bots stand in as opponents where humans are not available or desired. Alpha Go uses machine learning to master the ancient game of Go.  It is a testing ground for researchers to invent smart, adaptable algorithms to tackle problems. Social network services and algorithms are used to forecast box office takings of a motion picture.

Retail

In the retail sector, machine learning techniques select the product mix, dynamic pricing, promotions, and placement on sites, apps, or a store window. It increases sales by 10 to 20 percent and offers a better shopping experience. It helps to predict the stock for each season, hence eliminating waste across the supply chain, saving time and resources. For example, Walmart analyzed weather data and social media to stock more pop tarts than usual in hurricane-affected stores.

Apple, Amazon, Nordstrom with smart mirrors and beauty e-retailer, Birchbox, opened brick and mortar stores to provide an experience to their customers while collecting data to get insight and feedback to improve their products and services online. Exploration of trackable digital data in real-time helps in providing recommendations, matching styles according to preference, and persuades customers to purchase according to their taste even before consumer searches for an item. Retailers such as Stitch Fix offer services such as personal shoppers by matching people to products, shoppers to stylists to optimize the lifetime value of consumers and recommend apparel to shoppers.

Restaurant

At present most of the restaurant recommendation sites are review-based. A customer searches a search engine for local choice with countless results to sift through. Restaurant recommendation sites such as Luka and Nara mine the internet for information, do trend analysis, learn user’s restaurant preferences, and scans online reviews to recommend a place and not a list. Predictive analysis is also used to forecast annual sales of restaurants based on given objective measurements.

Hotel and Tourism

Artificial neural networks, Gaussian process regression, and support vector machines are the most widely used machine learning methods for forecasting in this sector. These techniques help in suggesting mid and long-term forecasts by evaluating demand with seasonal fluctuations. These approaches are also used in hotel site evaluation by designing automated web Geographical Information System (GIS) applications. Hotel location selector and analysis toolset predict various business success indicators associated with a location for example in Beijing, the tool calculates and visualizes desirable sites contingent on specific features of the hotel. For shorter forecast periods, methods help in improving standards with a direct approach and the introduction of tools to automate the information discovery process. The issue with these approaches is exploratory data analysis needs to extract information from seasonal, cyclical, and trend components. It is time-consuming and requires domain experts.

Banks

In the banking industry, self-improving algorithms detect fraud, content, and information extraction, establishing identity, efficient credit card transactions, correct decision-making and even predicting credit scores for ascertaining bad loans. According to IBM research, in the US, the cost of fraud is $80 billion dollars annually with $2.4 billion lost through credit cards.

Management of credit cards efficiently and effectively helps to uncover fraudulent transactions and immediately trigger a card rejection at a point of sales. For example, Feedzai utilizes machine learning models to detect fraud up to 30% earlier than traditional methods. Kabbage applies predictive analysis to offer investment to small businesses and consumers through automated lending platforms. LendUp uses algorithms to pinpoint customers who are most likely to repay their loans. Allowing data processing enhances decision-making for companies such as Zest Finance.

BillGuard applies models and techniques to alert users to bad charges. DataMinr helps in real-time information discovery and Alphasense employs algorithms for high-frequency trading by identifying true signals automatically among massive amounts of data. In 2014, more than 40% of new hedge funds used computer models for their trades. Moreover, consulting firm A.T. Kearney reckons that robot-advisors will handle up to $2.2 trillion by 2020.

Healthcare

In recent years, there has been an unprecedented rise in the availability and size of clinical data. This presents an opportunity to develop various approaches better personalized care. Similarly, dose-response models and disease progression models characterized by states that evolve over time help to treat individual patients best possible way. Pharmaceutical companies use data examination to deepen their understanding of diseases and design better products for treatments. Machine learning also helps regulators and policymakers to define policies that increase healthcare value and safety.

For example, Enlitic uses deep learning and image analysis to make diagnostics faster, more accurate, and more accessible. Ginger.io uses a combination of smartphone technology, data science, and clinical services to create a personalized affordable way to deliver mental health care. Medaware employs big data analytics and algorithms to cut healthcare costs by identifying and preventing prescription errors in real-time. And Lumiata applies real-time analytics to help hospital networks and insurance carriers in providing higher quality care to more patients in less time.

Law

Machine learning acts as a complement to legal counseling and helps lawyers save time and money by suggesting future legal outcomes against data from past client scenarios and other relevant public and private data. For instance, Supreme Court Ideal Point Miner is 76.9 % accurate for making predictions about US Supreme Court judge’s individual judicial preferences and voting behavior.  It also aids close the justice gap by offering digital advice to those who cannot afford a lawyer. It offers dispute resolution platforms to help mediate between users. It also cuts the amount of time invested in legal research. An example is the use of language analysis by Ravel to check a case, pinpoint a particular case it refers to, and figure out what ties the cases together.

Education

Machine learning provides new opportunities to tackle the challenge of large-scale grading automatically by analyzing data generated while students interact with teachers and other students.  Data mining and predictive analysis are applied to offer personalized feedback, track progress, and recommend steps for improvement.

Content analytics organize and optimize content and scheduling algorithms search for an optimal and adapted teaching policy to ensure help students learn more efficiently. It is also used for back-office operations by companies such as Edulog for bus scheduling, MyEdMatch, and TeacherMatch for matching teachers and schools, or Enterprise Resource Planning systems to predict enrollment, enhance security, or boost retention.

Government

Modern search engines and natural language processes are used to organize and explore public documents. Machine learning increases operational efficiencies by analyzing data sets and finding patterns and anomalies. Additionally, the identification of important trends and subtle yet complex patterns generate the most relevant and high-impact information. This information detects and prevents fraudulent transactions, provides efficient, effective public services, and helps public officials to make informed decisions.

Analysis of data helps in detecting illegal activities such as money laundering, criminal activities, trade of counterfeit items, or spread of terrorism.  Identifying trends and correlation among different groups stop these activities before it is too late. An example is to run buses according to citizens’ needs or when raining and not on fixed timetables. Also, it reduces duplicates in the database by recognizing two businesses with the same address and phone number.

Machine learning provides intelligent support to various aspects of the economy. Deducing from data is helping in improving productivity, daily working logistics, increase revenue, and better learn the ambiguities of the real world. Its uses include recommending and up-selling.

For instance, recommendations drive 50 percent of LinkedIn connections, 75 percent of Netflix views, and 35 percent of Amazon sales. It enhances customer satisfaction while reducing costs and consequently helping a business’s success. These techniques also make a government’s operations open, flexible, fast-paced with less bureaucracy and more direct input from citizens.

Tagged