Machine Learning Could Help More People Quit Smoking
Smokers may gain a heightened awareness of when they are going to cave to a cigarette thanks to machine learning. It’s no secret: nixing a smoking habit is tough. My great uncle took a trip to the hospital after setting himself on fire trying to light up, and he still smokes. However, not all hope is lost for my uncle and the other estimated 1 billion smokers out in the world. Recent research used an ML model to identify trends in smoking habits that could increase intervention and save the lives of smokers worldwide.
The verdict is pretty clear: smoking is bad for you. But, it’s also highly addictive. Chances are likely that you know someone who’s struggled with it or maybe you even have yourself. As with any addiction, there’s no shortage of things to try — clinics, support groups, and home remedies are widespread across America. From going cold-turkey to religiously sticking on nicotine patches, people are willing to try almost anything. So why do so many of them still smoke?
Researchers representing Infinia ML and Duke University, Dr. Hongteng Xu and Dr. Lawerence Carin, collaborated with researchers from Duke’s Department of Psychiatry and Behavioral Sciences to gain more insight on what drives the smoking craze. The study took the data from 42 smokers who pressed a button every time they had a cigarette for a week and framed it into a machine learning model, specifically, a time-varying semi-parametric Hawkes process (TV-SPHP) machine learning model. Snazzy.
The researchers modeled smoking as a “self-triggering process” by focusing on the “temporal dynamics” of smoking which essentially refer to the daily/weekly smoking patterns of an individual. For example, someone may be in the habit of having a cigarette before their commute every morning. Previous research has linked temporal dynamics with physiological cravings that are measured by a smoker’s plasma nicotine levels. Researchers believe the correlation between these two factors is what makes it so difficult for a person to fully quit. By collecting data of an individual’s smoking events over the span of a week, researchers were then able to use machine learning to construct a timeline of each person’s cravings.
But the ML model didn’t just create a timeline. It took the data a step further by learning from the smoking events logged at the beginning of the week to then make predictions on when the individual was most likely to smoke again in the future.
What does this mean for us mortals? Researchers hope that smokers who are more self-aware of their temporal dynamics will have a better chance of seeking intervention. So, imagine that the ML model did identify a craving pattern that occurred around the time of an individual’s morning commute. With a heightened level of awareness that this is a time of high-risk, one could feasibly take preventive measures like calling an intervention counselor to talk with on the commute, thus countering the urge to smoke.
Researchers also noted that the model was able to identify possible connections between temporal dynamics and nicotine metabolism to be explored in later research. Overall, the continued use of machine learning to understand more about how we function as humans on both a physiological and psychological level has the potential to have a profound impact on our lives. The CDC reported in 2014 that cigarette smoking is responsible for over 480,000 deaths per year in the United States, with total yearly economic cost exceeding $300 billion (Centers for Disease Control and Prevention (CDC), 2014). While this study was isolated to a particular group, it shows one way that machine learning could start aiding in the crusade to help improve and potentially save the lives of smokers around the world.