experiment
Our scores are rigorous and designed to mirror real-world utilization. We evaluated the forecasts for 100 randomly chosen stations for each 30- and 60-min intervals, sampling occupancy 48 occasions each day (each 30 min) over per week.
This mannequin was benchmarked towards a really robust baseline of “keep present”. This baseline merely assumes that after a sure variety of minutes (H) sooner or later, the variety of accessible ports will probably be precisely the identical as the present quantity.
Though easy, this baseline may be very tough to interrupt, particularly over quick intervals of time. For instance, our knowledge reveals that on the US East Coast, not more than 10% of ports change availability state inside half-hour. Since more often than not the state doesn’t change, the best prediction (no change) is sort of all the time right, making the duty of including predicted values very tough.
We targeted on two essential metrics to measure the accuracy of the mannequin for predicting the precise variety of free ports: imply squared error (MSE) and imply absolute error (MAE). MSE/MAE ≥ 1 The free ports ratio measures the accuracy of an important binary activity for customers: “Can you discover not less than one free port? (Sure/No).”
