It switches itself on and off again

SEP 08, 2021

Lucien de Voux

Machine learning (ML) is defined by McKinsey as the gaggle of algorithms that learn from data without relying on rules-based programming and that have the endless patience and capacity to munch through vast, unimaginable quantities of data to find significance, insight and information. It’s also an opportunity estimated to be worth nearly $6 trillion, with IDC claiming the market to be worth around $500 billion by 2024. ML is making inroads into every industry and sector and changing them in ways that many people don’t realise.

One of the most unexpected ways is in spamfiltering. Yes, all that intelligence and algorithmic wonder channelled intothe mundane task of ensuring that spam is classified properly and thatpotentially risky emails are removed. Considering that the average personreceives around 83.6 emails a day, according to EmailAnalytics, that’s a staggering sum of just over 30 000 emails a year. Global spam volumes accounted for 45.1% of email traffic as of March 2021, so ML that removes spam from the inbox is a welcome gift.

It’s equally of value in autocorrect, virtual assistants, intelligent facial recognition, financial market management and fraud detection, and in the ubiquitous chatbots that have helpfully replaced humans on the Q&A frontlines. ML has also transformed how certain healthcare practices and processes have evolved, shifting patient care even further into the forefront while helping physicians to reduce their admin workloads and minimise potential errors. In fact, it’s in radiology where AI and ML have excelled – catching fragments and potential problems in scans and alerting physicians at speed, helping them to prioritise patients accordingly.

Machine learning has inserted itself into almost every area of the business and has proven its value across most sectors. In retail, machine learning is emerging as a tentative chatbot success story, but a definitive value-add in improving customer experiences and relationships. In the industrial sector, it has helped organisations to make granular changes to systems and approaches that have saved money and improved success parameters over the long and the short term. And the use cases evolve with need, sector and application.

Moving towards a more data-driven organisation and leveraging the power of ML can be expensive if done haphazardly.

Jon Jacobson, Omnisient

The value of ML lies not just in its ability to learn patterns of behaviour, or to deep dive into data, but in helping organisations to actually discover the value hidden with their data. Yes, this is a tired old data trope that’s trotted out with AI and analytics, but it’s a tired old cliché for a reason – around 80% of data is lost to the average business, says McKinsey. This means that they’re not even close to knowing what data they have or how to use it. ML can be a business ally, as useful as the latest investor or smart stakeholder. But, there’s a caveat: it’s just a technology and its value lies in its implementation, use case, capability and relevance.


Leveraging machine learning for value.

Brainstorm: How can the organisationfully leverage ML to achieve more today and in the future?

Jon Jacobson, co-founder, CEO and CTO, Omnisient: They need to understand the problem they’re trying to solve and whether they have the correct data for it. Moving towards a more data-driven organisation and leveraging the power of ML can be expensive if done haphazardly.

Hanno Brink, machine learning engineer, Synthesis Technologies: Selecting the right tools can be a huge challenge, so it’s important to keep flexibility in mind.

Sarthak Rohal, VP: IT Services, AlphaCodes: Organisations need to embed AI methodology in their end-to-end business model, which combines the human capacities for learning, perception, and interaction, all at a level of complexity that ultimately supersedes our own abilities.

Fred Senekal, head of R&D, Learning Machines: In order to fully leverage machine learning, organisations need to become significantly more data-driven. Very often, this requires a culture of empowered employees with the right access, knowledge and tools and a leadership that makes it a reality.

Brett St Clair, CEO, Teraflow: Ask the right question and then find out what data you want to use to answer that question. All of this is about data that informs decisions.

Brainstorm: What are some of the standout ML solutions, approaches and developments right now?

Chris Cooper, general manager: ISG MEA, Lenovo: In the past decade, the cost of full genome sequencing has become more affordable as high-performance computing has become more attainable. Scientists previously could only sequence about 2% of genomic data, but they can now look at the entire genomic sequence of thousands of families at once. This progress can be the key to more effective discovery of genes that cause disease or the development of precision medicine.

Yaron Assabi, CEO, Digital Solutions Group: While some sectors, like retail, had to evolve or be disrupted given the pressure to digitally transform at a rapid pace, others had to adopt a defensive strategy, as consumers changed their buying behaviour and migrated online. It’s here that machine learning has played a critical role over the past year alone as it has enabled improved processes, enhanced customer experiences and enabled intrinsic personalisation.

Shakeel Jhazbhay, general manager: Digital Business Solutions, Datacentrix: Cybersecurity applications: these have become hugely important in terms of remote working, particularly in terms of managing the volume of transactions and the accuracy of incident reporting. Business forecasting and reporting: analysing data to help reduce uncertainty, anticipate changes in the market and predict future developments, and improve business decision-making.

Reven Singh, sales engineer, InterSystems: Everyone is already experiencing ML in their everyday life, from helping virtual personal assistants understand our speech, such as with Amazon Alexa and Apple Siri, to spam filters and malware detectors. Think of how Facebook suggests new friends and new groups to you; that’s using ML.”

Riaan Devilliers, business analyst, LAWtrust Information Security: Netflix is the world's leading internet stream service with 160 million customers worldwide. Some analysts think it is Netflix's early adoption of ML that made it the world leader.

Brainstorm: What would you define as best practice in implementing or investing in ML today?

Mandla Gqada, solutions architect and engineering lead, MakwaIT: Machine learning experts across the different divisions in an organisation, instead of having one central, isolated machine learning team. This will enable machine learning experts to work side by side with domain experts who understand the data better than the machine learning experts.

Marilyn Moodley, country leader for South Africa and West, East, Central Africa, SoftwareONE: CIOs should push to empower machines to do more, better learning ahead of the task. This requires rethinking on how machines take in data. Businesses should not think of themselves as a collection of tasks, but, rather, view their operations as brought to life by streams of data that run through workflows made up of those tasks.

Craig Stephens, advisory business solution manager, SAS in South Africa: With so many different approaches, models, and methodologies to choose from, each company’s ML journey will be guided by its strategic imperatives. But it’s still worthwhile to build simple, white-box models using regression and decision trees. Simpler models are also easier to deploy, which makes the IT and systems operation teams happy.

Nkosi Kumalo, managing executive, BCX Exa: Without a clear understanding of what you want to achieve, it’s impossible to measure success. This includes identifying the opportunities and defining the use cases. From a best practice perspective, there needs to be a consensus based on IT fundamentals, but the specifics may vary depending on the technology stack used to execute the ML initiative.”

Machines putting patients first

How Palindrome Data leveraged machine learning to predict retention and viral suppression in HIV treatment.

There are two things that the current South African healthcare sector knows are true. The first is that South Africa has one of the largest HIV epidemics in the world, with more than 7.5 million people living with HIV; and that access to data, and use of this data, is limited by rural locations and limited access to healthcare facilities and technologies. That said, thanks to the hard work and commitment of government agencies, universities and various funding organisations, there’s a significant quantity of accumulated data that has the potential to be used intelligently to help practitioners make real-time, action-based decisions that put the patient first. This is where machine learning and Palindrome Data step in.

“What we did was take the data and use machine learning to build predictive models so we could build tools and job aids for clinicians and frontline healthcare workers to make better patient decisions,” says Lucien de Voux, director of Market Strategy at Palindrome Data. “It’s essentially the use of machine learning on healthcare data to help clinicians understand the patients most at risk.”

Ask the right question and then find out what data you want to use to answer that question.

Brett St Clair, Teraflow

Accessing the data was the key part of solving the problem and the team was fortunate enough to have contacts at universities such as Wits and the National Institute for Communicable Diseases and other academic institutions and establish partnerships where they got the data, and provided value back by developing tools and providing insights.

“At first, it was very much a research initiative, but over the years, we’ve built up evidence and published papers that have allowed us to be on the ground and deploy models through tools and job aids,” says De Voux. “What we’re doing now is going beyond the theory and taking the tools to clinicians and facilities on the ground to improve usability and patient engagement by leveraging physical tools.”

Predictive algorithms

A lot of machine learning goes on in the background as the company takes the models and builds both digital applications and paper-based tools around the models. Having paper-based solutions was critical as many clinics are remote and have low resources, running exclusively on paper-based systems. The machine-learning developed tools have been translated into a paper format so they can be deployed into clinics in remote areas.

“One of the problems facing clinicians when it comes to HIV is retention,” says De Voux. “HIV can be well managed as long as patients stay on their treatment, but up to one in three patients in southern Africa drop out of care, so the problem is not knowing who drops out and who stays in. Our predictive modelling and machine learning take these big data sets to build predictive algorithms so we can triage which patients are at high risk of lost follow-up, of stopping treatment. When clinicians know which patients are going to have trouble staying in care, they can shift their resources to those patients, which delivers impact and cost savings.”

Palindrome Data has published several papers built around the machine learning and predictive outcomes of its work, and has an accuracy of three out of four for viral load suppression, and an accuracy of two out of three for those likely to drop out of care.

“When you can make those predictions based on the data, then you can change your intervention strategy as opposed to retroactively finding patients. It goes beyond just identifying patients at risk, but into tailoring solutions and using machine learning and big data to better engage with patients and deliver personalised solutions,” says De Voux.