Machine Learning Reveals the Unexpected

With a platform as versatile as Stratifyd’s, it’s important to understand that there are many uses for Big Data and powerful analytics tools. Businesses use our AI powered machine learning to analyze their customer experience or better understand their employees. Companies look at product management or try to understand what might happen in the future. Within all collected data, there are multiple layers of information and various ways of interpreting that information to gain better insights. Stratifyd wanted to demonstrate the versatility of its product by assessing fatal automobile accidents in the State of North Carolina between 2010 and 2014. Our analysis of traffic accidents demonstrates the various ways that data can be layered, interpreted, and understood. Our results also demonstrate that sometimes, unsupervised machine learning can reveal insights that go against expectation and reveal information previously unforeseen.

Stratifyd examined information from the North Carolina Department of Motor Vehicles and analyzed nearly 6,000 accident reports. Our findings uncovered a great many details that run contrary to what many consider “common knowledge.” Here are a few of our results.

State roads, generally small two lane highways, accounted for the largest percentage of fatal accident locations, accounting for 44% of fatal crashes statewide.

• Wet or snowy roads only accounted for a mere 15% of fatal accidents, statistically refuting the idea that more fatal crashes happen in inclement weather.
• Lane Departure without signaling accounted for 55% of fatal accidents analyzed.
• Only 30% of fatal car crashes were the result of drivers exceeding posted speed limits.
• Pedestrians were involved in 13% of fatal accidents between this time period.
• Fatal accidents on Interstate Highways were only 9% of reported crashes.
• Drivers between the ages of 16–20 accounted for only 10% of fatal accidents across the state, contrary to consensus that younger drivers are more accident prone.
• Drivers between the ages of 50–65 accounted for 25% of fatal crashes, indicating a higher likelihood of a fatal crash at older ages.

Already, the findings present some unexpected insights that go against traditional thinking on fatal car crashes. By utilizing unsupervised machine learning, Stratifyd allows our robust artificial intelligence to analyze the data automatically, searching for patterns and trends to return the true picture of analysis. Now that we’ve looked at some extenuating factors that affect car accidents, let’s look at some other common concepts.

• 55% of fatal car crashes statewide involved alcohol, upholding the commonly held belief that drunk driving causes fatal accidents.
• Just 12% of fatal accidents involved a motorcycle, potentially indicating that a crash on motorcycle is not an immediate death sentence.
• Only 6% of fatal accidents involved a semi-truck.

The time of day is generally a factor, with most believing that more fatal accidents happen at night than in the day. However, 55% of fatal accidents happened during daylight hours, with 37% happening on a dark, unlit road; and 11% happening at night on a well lit road. Additionally, the seasons play a factor in car accidents as well, with the most accidents coming in the Springtime.

Comparing major cities in North Carolina, our analysis revealed that Charlotte, NC is the most dangerous city in which to drive for the state. It accounted for 48% of fatal car crashes in the time period assessed. Raleigh and Fayetteville came in second and third respectively with 27% and 25% of fatal accidents in major cities.

With a robust data analytics platform that uses machine learning and natural language understanding to analyze raw data, companies will reveal insights they were not expecting to find. The results of this study show that some common ideas about fatal accidents are misconceptions. By allowing artificial intelligence to analyze data without human interference, a wide variety of surprising information can come to light. Analysts can then use this unfiltered information to augment their deep understanding of whatever it is they are trying to assess in order to achieve pure actionable insights.

Originally published at www.stratifyd.com.