Why You Must Twist Your Data Scientist's Arm To Estimate AI's Value
How does the New York City Fire Department utilize predictive models in their operations?
The New York City Fire Department (FDNY) utilizes predictive models in their operations to prioritize which buildings in the city are at the highest risk of having a fire12. They use a data tool called FireCast 2.0, which applies an algorithm from five city agencies, taking into account as many as 60 different risk factors such as a building's age, whether it was in the middle of foreclosure proceedings, or had active tax liens. This system has simplified the process of targeting the most fire-prone buildings for inspections, many of which hadn't been inspected in years. Since its deployment in 2013, more than 16% of the city's fires were in buildings that had been inspected in the past 90 days, suggesting that the right structures were being prioritized for inspections. The FDNY is also working on upgrading to FireCast 3.0, an even more powerful tool that will analyze three years of data from 17 different city agencies for every one of the 330,000 buildings in the city, assigning each building a fire risk score1.
What is the primary function of predictive AI in enterprise applications according to the article?
The primary function of predictive AI in enterprise applications, according to the article, is to serve as an antidote to information overload. It helps businesses prioritize and make data-driven decisions by providing predictions based on historical data and patterns. For instance, predictive AI can prioritize prospective customers based on their likelihood to buy, or it can be used to identify high-risk buildings for inspection based on historical fire hazard data. The use of predictive AI allows for more efficient and targeted decision-making, leading to improved business outcomes.
What does an AUC score of 0.83 indicate about the quality of a predictive model?
An AUC score of 0.83 indicates that the predictive model has very good performance. AUC stands for Area Under the Curve, and it is a metric used to measure the performance of a binary classification model. The curve referred to is the Receiver Operating Characteristic (ROC) curve, which shows the trade-off between sensitivity and specificity at different thresholds.
An AUC score ranges from 0 to 1, with 0.5 being the score for a model that is no better than random guessing, and 1 being the score for a perfect model that can accurately classify all cases2. Generally, an AUC score of 0.9-1.0 is considered excellent, 0.8-0.9 is good, 0.7-0.8 is fair, and 0.6-0.7 is poor. Therefore, an AUC score of 0.83 is considered good and indicates that the model is significantly better than random guessing and has a high probability of correctly classifying the classes.