Machine learning has invaded and challenged traditional methods in unimaginable ways in the recent past. Continuing in the same vein, today we look at some of the innovative ways machine learning is addressing alcohol drinking related issues. Be it drunken outburst on social media or drunken driving, new age algorithms are preparing to help you save both your life as well as your reputation.
Detecting outbursts on social media
More than harmful, drunk tweeting or social media updates are regretful. And a group of researchers at the University of Rochester have been building a machine learning tool just to address this issue. Using geo-tagged tweets of about a year’s duration (2013-14), the team was able to train a dataset to identify drunk tweets with an accuracy of over 90%. The team employed SVMs to classify tweets based on questions considered typical of alcohol induced behavior. The only caveat is that this algorithm doesn’t prevent you from drunk tweeting yet. Twitter might have to do the thinking on that.
And while twitter is yet to recognize the utility of such a tool for R&D investments, Facebook has already done that. Almost a year ago, its AI Department chief Yann Lecun expressed his intent of building a virtual assistant that could help individuals not post images or text they would regret later. Arguably, it would be based on far more comprehensive deep learning techniques. An example of such a tool could be the virtual assistant identifying the key facial traits of your drunk selfie, matching it with your sober looking photos to determine that you are indeed heavily drunk and then warning you not to post it publicly. Sounds helpful indeed!
Detecting drunken driving
How do cops detect drunken drivers? Detecting a drunken driver is a tough job and currently based on using cues that have had past correlation with success. Vehicle movement, time of the day, people in the vehicle, location, past experiences, etc. all play a major role in helping a cop identify which driver should be tested with breath analyzers. But this requires extreme attention. Needless to say, some help in the form of automatic detection will be of great help. Also, if we see a future of self-driving cars, a single drunk driver on the roads can push such plans astray.
It is with this intent that a research team at University of Dayton Research Institute is working on the Traffic Camera Distracted Driver Project, a machine learning approach to detect vehicle co-ordinates, anomalies in vehicle path and use the same to classify drunken driving behavior. While this research is in early stages only, hopefully some good will come out of it really soon.
In addition to these, other tools have also been built to address the issue. Japanese watchmaker Tokyoflash had launched a smartwatch that uses a blend of two predictive tests: an outcome of a gamification rule and a digital breath analyzer tool to predict your blood alcohol levels. Along with that, researchers are also working on building a digital drunk detection tool in cars using techniques such as auto breath analyzer or infrared finger touch tests. Results of such tools is planned to be linked with the car’s ignition system with conditional access.
Whatever the methods be, we eagerly look forward to a future free of stupid actions borne from drunkenness.