Weather condition and the type of Crime Types of Correlation
The skills we demoed here can be learned through taking Data Science with Machine Learning bootcamp with NYC Data Science Academy.
1. Contributed by Brian Lam.
2. Student of Data Science with R: Data Analysis and Visualization with Paul Trowbridge in Summer 2015.
3. The post is based on using data analysis and visualization to investigate weather condition and the type of Crime occur in San Francisco.
Background
Does criminal prefer to work on specific weekdays or temperature? Does Burglary or theft choose sunny day over rainy day? Tenderloin area is infamous of high crime area, what type of crime do they have the most? Does criminal prefer to work on night time? My study of the Crime tries to find answer from San Francisco Police crime. I found San Francisco Police report from Kaggle website and download weather report using weatherData library. My study will be target 9 types of crime included arson, assault, burglary, DUI, Drug, vandalism, vehicle theft, robbery and bribery.
Objective
1. Using San Francisco Police Report and aggregate type of crime, hour of crime, and weekdays of the crime. Using WeatherData library to retrieve San Francisco weather from December 13, 2014 to May 13,2015 and aggregate the temperature and raining condition by hour. Merge weather data with crime data.
2. Create a visualization map to find crime pattern at different crime at San Francisco, graphing each crime by hour to visualize highest crime by hour, graph each crime per weekdays to visualize and highest crime of the weekdays.
library(weatherData) library(plyr) library(reshape) library(RgoogleMaps) library(ggplot2) library(lubridate) library(corrplot) library(grid) #Get the weather data in San Francisco from 2014-12-13 to 2015-05-13 d3<- getWeatherForDate("SFO", start_date="2014-12-13", end_date = "2015-05-13", opt_detailed = TRUE, opt_all_columns = TRUE) temp <- d3[, c(1,2,3,13)] temp$Time <- format(as.POSIXct(temp$Time, format = "%Y-%m-%d %H:%M%S"), "%m/%d/%Y") # Coerce the string to PoSIXIt date temp$TimePST <- format(as.POSIXct(temp$Time, format = "%Y-%m-%d %H:%M%S"), "%H") temp$Time <- format(as.POSIXct(temp$Time, format = "%Y-%m-%d %H:%M%S"), "%m/%d/%Y") # Convert weather Condition to 1 for Rain and 0 for Clear or Cloudy day for (i in 1: length(temp$Conditions)){ if(temp$Conditions[i] == "Light Rain" || temp$Conditions[i] =="Heavy Rain" ||temp$Conditions[i] =="Rain") {temp$Conditions[i] = 1 } else { temp$Conditions[i] =0 } } #Aggregate the hour of the date for weather in San Francisco temp$TimePST <- as.integer(temp$TimePST) temp$Conditions <- as.integer(temp$Conditions) temp <- ddply(temp, c("TimePST", "Time"),summarise, Temperature = mean(TemperatureF), Condition = mean(Conditions) )

Using RGoogle Map library to plot the the location of crime base on latitude and longitude.
Tenderloin (area in AT&T park) is infamous for highest crime in the city of San Francisco but not all crime are the same. Theft crime is mostly occurs at this region but other type of crime are even distributed among San Francisco.
From this chart, we can see temperature has negative correlation with the crime. Most criminal prefer to do the crime at cooler temperature than hot day. However, assault rate is higher at hot temperature than cooler temperature. As expected, bribery and DUI has no influence by the change of temperature.
Most crime has no significant difference between raining day and sunny day. It is interesting to note more assault and vandalism occurred on raining day than non raining day. As expected, bribery, DUI and Drugs related crime does not depend on raining condition. Most Robbery occurred during non raining day.
Interestingly, we see that the average rate of crime are different per weekday. Assault rate is highest on Friday but dropped significantly on Saturday. Burglary is highest on Friday and Wednesday but low on Sunday. Robbery is highest among the week. Vehicle theft like to steal car on Friday. As expected, DUI is higher on Friday and Saturday.
Most
Criminal are most active at night time from 4pm to 12 am and lowest at 3-6 am. Most interestingly to note highest drug related arrested at around 4pm. DUI related arrested is mostly around 11pm to 12 am.
Conclusion
Contrast to our belief, most criminals do not prefer to do their crime at sunny day over rainy day. From our correlation plot, we found criminals are most active at cooler temperature than hot temperature. However, San Francisco has moderate temperature swing, most of the time San Francisco temperature range is from 60-80s. I believe if I do a research on temperature of New York to temperature, criminal activities will decrease when the temperature drop 30 or less. The next step of this research is to create a predictive model to predict the type of crime base on the temperature and weather condition of the day.