Machine learning based on traffic police statistics. Any ideas?

Question

Machine learning

I want to practice machine learning. In mind there is open traffic police data. In particular, several hundred thousand registered accidents over the past three years. From the data there:
Table of accidents accident
id
region
address
geocoordinates
date
time
type of accident
cause of accident
number of vehicles
number of injured
number of killed
sex of all injured-killed
weather
condition of coverage
table vehicles
class
brand
model
color
accident id
drive
year of manufacture of
damage
Tell me, how can you exercise on this data in terms of machine learning / neural networks? And what algorithms are suitable for this.

Answer 1

You can publish the results of the study in magazines, media and entertainment venues. Whatever, but the result will be, at least the awareness of the population, which is more afraid of terrorist attacks than driving or walking down the street (where the probability of dying is higher than from terrorist attacks).
If I had time, I would at least make a heat map with the most emergency places in the city with reference to the time of day and seasons, and put it in the public domain.
It will be difficult to automatically analyze the cause of an accident, you need to at least categorize the cases (drunk driving, fell asleep, did not see, did not have time).
It makes sense to consider the make / model if there are the results of the accident (alive / dead / disabled - with the degree of each case), and the details of the accident (with whom it crashed, at what speed relative to each other, with what mass, etc.) otherwise, write that the Ford is unsafe, although in the statistics a truck crashed into it.
I do not advise you to classify by gender, they will sue you :-)
To choose an algorithm, you need to understand the goal, in what form you expect the results of the work. He will not be able to tell you "put asphalt over there, ban such people from driving, stay at home at 22:00".

Answer 2

nothing new except for emergency sections was invented
especially funny, how do you propose to practice with the brand of the car - well, make a selection for buyers which brand is "olpasnaya", although all cases are independent

Answer 3

There is not enough data for training.
Only perhaps some kind of analysis of statistics such as a heat map.