Implement demand prediction

[erikmoh]

• check Odd and Erling method and solutions
• evaluate simple method vs their results
• evaluate effect of optimal prediction

simple method:

• count incidents per grid, and then sum area of grids nearby ambulance

• divide into base station responsibility district

• count number of incidents for each district for each hour only

• create a dict with district1: {00h: 5, 01h: 9, ...}

• dispatch penalty based on how many incidents in district that hour

• potential to expand into district1: {month1: {day1: {00h, 01h}, day2: {00h, 01h}, month2: {day1: {00h, 01h}, day2: {00h, 01h}}

advanced method:

• neural net
• input: section/grid, month, day of week, hour, (weather)
• output could be all grids/section or for only one
• experiment less features: only weekdays, time of day insead of hour
• check preprocessing
• try classification: 0, 1, 2, 3 or more
• poisson regression