The Advantages of Using ML for Behavioral Analysis in Health Care, IT Security, and More.
In 2019, Yan et al. published a study exploring the suitability of machine learning to analyze online learning behavior for predicting student performance.
The time was ripe for such a study, the authors argued, thanks to a recent explosion of learning-related data from which to derive insights. The data increase was due in part to the popularity of MOOC and similar platforms (indeed, the increased popularity of online learning due to the pandemic will no doubt spurt further studies in this regard).
The study authors ran data harvested from 192 student records in online learning platforms through an artificial neural network (ANN). The authors included data points such as student age, gender, connection time, hits count, and days of access. From this, researchers were next able to help teachers roughly predict performance by analyzing students’ behavior online, finding that days of access correlated the closest to student performance.
Despite using a relatively small dataset – the authors themselves described it as “limited” – they could also provide real-world suggestions for teachers based on this behavioral analysis. “If the teacher finds that some students have a much lower number of days of access and hits count than the average number on the learning platform,” they wrote, “he or she can pay appropriate attention to them and give them effective learning support to prevent them from failing the exam.”
The above experiment is just another example of how machine learning (ML) and other forms of artificial intelligence (AI) are analyzing and helping to solve problems in all aspects of our lives. From municipal authorities using ML models to design smarter cities, to health-care professionals analyzing big data to personalize health treatment, to researchers using natural language processing to discover new drugs, ML is suddenly everywhere.
One of those areas is behavioral analysis (and even behavior modification through such analysis). But before we explore how ML can achieve this, it’s worth taking a look at what behavior is in the first place.
What exactly is behavior?
Behavior is a set of actions or mannerisms made by animals and humans concerning their environment and themselves – whether it’s stress sweating before a job interview, or jumping up after receiving a fright. Behavior can apply to systems and artificial networks, as well. Behavior can be voluntary or involuntary (and, by extension, conscious or subconscious) and is essentially the response of an organism to any number of stimuli.
Because behavior consists of “any observable overt movement of the organism generally taken to include verbal behavior as well as physical movements,” it is by nature an observable physical activity.
How does ML analyze (and even help modify) behavior?
Machine learning involves the automatic analysis of behavior-related data – in most cases, too much or too complex data to analyze using traditional methods – and takes many forms, including supervised learning, unsupervised learning, and reinforcement learning. ML models learn from data (and improve their effectiveness from training using relevant datasets). Once trained to an adequate level, can spot patterns or make decisions autonomously – often with a degree of accuracy on par with humans.
Behavioral economics and consumer psychology researcher Ahmad Tanehkar quotes Daniel Kahneman as saying, “We (humans) are pattern seekers,” and the same thing applies to ML models. ML models automatically identify patterns within data, then use those patterns to predict future actions or behavior.
ML models for specific use cases are trained using various dataset types, including health data, financial data, social media data, or data captured by internet of things (IoT) devices and other sensors.
The importance of sensors for behavioral analysis
For ML models to observe and analyze the physical behavior of organisms such as people or animals, they need the right tools, says Enrique Garcia-Ceja, a researcher at Norway-based SINTEF Digital. Humans and animals have their senses to do this for them. Machine learning models, lacking eyes, ears, and noses, must rely on artificial sensors such as microphones, thermal and RGB cameras, temperature sensors, vision and imaging sensors, or vibration sensors.
Luckily for ML researchers, advances in sensor technology have led to various wearable technologies outfitted with sensors. Smartwatches and fitness trackers are one example, but developments in flexible and hybrid electronics have also resulted in various other types of wearables, including vital sign-measuring shirts and self-heating smart jackets.
All these types of products are the new eyes and ears of ML models used for behavioral analysis.
Applications of ML for behavioral analysis
We led off this blog talking about using ML to analyze and draw conclusions from online learning data, but that’s not the only behavioral analysis application that’s improved through ML models.
Health care, physical health, and personalized medicine
The wearable technology we mentioned earlier is used every day by individuals to analyze their behavior – and, if necessary, make changes to benefit their health. Handel et al. say wearables – including everything from smart fitness trackers, to sensor-equipped yoga pants, to smart helmets – can be used by individuals to “understand and make a personal plan to change their health behaviors (e.g., exercise, eating, sleep).”
The application of AI and ML algorithms to wearables data, in turn, provides real-time insights to make sense of all that data and help improve the health and fitness of the user.
ML provides similar behavioral analysis benefits on the health care front. Health-care professionals increasingly rely on medical-grade wearable devices such as ECG monitors, blood pressure monitors, and biosensors – all of which help practitioners monitor patient health and establish baselines remotely, a massive bonus during the Covid-19 pandemic.
Health-care applications also include ML algorithms in personalized medicine, using predictive analytics to devise ultra fine-tuned treatment regimes based on a patient’s unique data. Most ML models in this area involve supervised learning and can be used to treat particularly damaging behavior, such as addiction, or monitor for potentially life-threatening condition changes in a machine-learned baseline in patients with Covid-19.
ML can also make a big impact on the behavioral analysis of potential insider threats (among employees, contractors, or third-party vendors) within an organization’s corporate systems. Insider threats are especially insidious because they’re harder to detect using traditional security measures such as firewalls and intrusion detection. It’s a reason for the rise of Zero Trust architectures, whose effectiveness has been aided exponentially by applying ML models to an organization’s security posture.
Indeed, monitoring user activities to identify behavioral anomalies through user activity monitoring (UAM), user/entity behavior analytics (UEBA), data loss prevention (DLP) and other techniques have become more common as insider threats have grown. These techniques are primarily necessary because of the sheer scale involved as data volumes grow within the enterprise. “Even if you knew what to look for, finding anomalous behavior then connecting the dots to develop a complete picture from a huge number of activities may turn out to be humanly impossible,” says IT Security Central, “especially if you have a large group of users.”
ML models have also aided the understanding of behavioral science in the economic realm. Researchers such as Daniel Kahneman, who we mentioned earlier, and Richard Thaler have shown that understanding the context of a person’s decision-making process can help models understand human motivation. ML models can use analytics to learn an employee’s behavior and send technology-induced “nudges”, which are essentially timely stimuli designed to spur individuals to better performance – an application known as ML-aided Nudge Management.
“The interactions between machine learning and behavioral economics can be mutually beneficial,” writes the behavioral economics researcher Tanehkar, mentioned earlier. “On the one hand, ML can be used to mine a broad set of data and find the behavioral-type variables that contribute to the emergence of different behaviors. On the other hand, ML algorithms that are embedded to identify biases and wrong assumptions would reach higher performance.”
How CapeStart can help
CapeStart provides data annotation services, pre-trained datasets, pre-built ML models, and AI/ML model development for behavioral analysis across various industries and applications, including health care, IT security, finance, and legal. Our integrated team of data scientists and ML engineers work with large organizations every day to improve and fine-tune MLl modeling and effectiveness, facilitating efficiencies and better business outcomes for our clients.