Drawing business conclusions

In this section, we'll apply our DNN model and understand how many times the BigQuery ML model is able to predict a rental duration close to the actual one.

We'll add a parent SELECT COUNT statement to the previous query to count how many times the difference between the actual duration and the predicted one is less than 15 minutes.

Let's execute the following query to calculate how often the trip duration predictions are far from the actual values:

SELECT COUNT (*)
FROM (
SELECT
   tripduration as actual_duration,
   predicted_label as predicted_duration,
   ABS(tripduration - predicted_label) difference_in_min
FROM
  ML.PREDICT(MODEL  `11_nyc_bike_sharing_dnn.trip_duration_by_stations_day_age_relu`,
    (
    SELECT
          start_station_name,
        ...