Machine Learning is an umbrella term for a set of tools which can be applied to solve so many problems in a variety of fields from computer vision and linguistics to logistics optimisation and fraud detection. Below are just few examples.
Waiting for equipment to break before fixing it can be disruptive and expensive. On the other hand, excessive maintenance is wasteful. Optimising when and how often to perform maintenance can save huge amounts of money and eliminate potential disruptions. This can be achieved by installing sensors into equipment and feeding the data in machine learning algorithms to perform ‘predictive maintenance’.
Example: Predict when equipment in a factory will fail, and hence when they need to be serviced.
Demand for products and services often fluctuates with factors such as seasonality, competition, or economic health, making it difficult to efficiently match capacity. Some of these factors are intuitive and their patterns can easily be predicted, whereas others are too complex or vary too rapidly to act on. Machine learning algorithms are capable of identifying patterns in data, where increasingly complex correlations simply require more data rather than more domain-specific expertise.
Example: Automatically schedule the ordering of ingredients to a restaurant based on historic consumption in order to minimise food wastage.
Receiving large volumes of customer interactions can overwhelm limited customer support services. Being able to organise messages by importance would enable staff to concentrate on the most pressing issues. By using natural language processing, a machine learning algorithm can be trained to recognise the importance and urgency of messages based on previously tagged messages.
Example: A letting agency would prefer to respond to a boiler break-down faster than a general query.
Real-time messaging platforms which allow images are open to abuse. This is difficult to moderate due to privacy concerns and the shear volume of messages. The ability to automatically flag offending images based on some predefined criteria can lead to a safer environment. This can be achieved with machine learning algorithms trained to classify abusive images. These images can be either be flagged for review, or blocked entirely.
Example: Flagging inappropriate images for review that are sent on a messaging platform for an online tutoring service.
Scanned Document Analysis
Manually extracting information from scans of documents in order to input them into a digital system is a common but tedious, time-consuming, and error-prone task. An automated process to extract and upload key information from paragraphs, tables, and key-values pairs has the potential to generate huge value. Using machine learning tools such as optical character recognition (OCR), document classification, and natural language processing (NLP), it is possible to perform this task much faster, cheaper and more accurately.
Example: Validating proof documents for a car insurance company