Billboard Selection - Google API

Billboard Selection Problem

Context

• Client is a major hospital, specialized in cardiac care.
• Client intends to build 4 billboards and place them at optimal locations in Bangalore City.
• Focal city chosen by our team is Bangalore.

Decision Problem

• Client needs help in identifying 4 potential locations to place billboards that maximize viewership of Hospital information subject to possible budget constraints.

Research Objective

• Identify the key entities that will help maximize the viewership of billboards.
• Within the Bangalore City, narrow down to major areas where billboards could be placed.
• Further narrow down to 4 optimal locations among st the major areas identified.

Why Bangalore City?

• With the raise in prevalence of heart attacks from 2% in 1960s to over 12% in 2008, Cardiac ailments is the most critical problem to solve.

• Cardiac Care is a critical specialty which requires sophisticated equipment and investigative facilities such as Echocardiography with Colour Doppler, Nuclear Scanning, Cardiac CT Scan, Cardiac MRI and Coronary Angiography.

• Highly specialized Doctors and Medical professionals are one of the key requirement of a successful Cardiac Hospital.

• Considering the needs to run a sophisticated Cardiac Specialty Hospital, Bangalore is an obvious choice. Since Bangalore is home to highly skilled Cardiologists and Medical Professionals, availability of research institutes in Bangalore helps in research needs of a Cardiac Hospital.

Entity List - Selection

To derive the Entity list, our target segment lie under:

• Middle Class
• Upper Middle Class
• IT Employees
• Age group (35-80+), which includes age group who may need cardiac treatment as well as the group that might want the cardiac treatment for their parents

Considering the target segment as well as the probability of viewership, below are the entities we have narrowed down to:

1. Bus_station
2. Train_station
3. Malls

Initial Analysis

Initial R Code:

#### INSTALL required packages if they are not installed already      #######################

require("RCurl")||install.packages("RCurl")
require("jsonlite")||install.packages("jsonlite")
require("plotGoogleMaps")||install.packages("plotGoogleMaps")
require("geosphere")||install.packages("geosphere")

#### LOAD required packages    #######################

library("RCurl")
library("jsonlite")
library("plotGoogleMaps")
library("geosphere")

#####   Google API KEY    #######################
key = "AIzaSyBcIbEaJzu2xrMtLSWo4jIealZ27E6dXNI"

#####   Loading the data for the required Places    ####################### 

# Bus Stations search
url = paste0("https://maps.googleapis.com/maps/api/place/radarsearch/json?&query=bus+stations+in+banglore&types=bus_station&location=12.9716,77.5946&radius=50000&key=",key)
doc <- getURL(url)
x <- jsonlite::fromJSON(doc)
busStations = x$results$geometry$location
#head(busStations)
busStations$type = "BusStations"

# Train Stations search
url = paste0("https://maps.googleapis.com/maps/api/place/radarsearch/json?&query=train+stations+in+banglore&types=train_station&location=12.9716,77.5946&radius=50000&key=",key)
doc <- getURL(url)
x <- jsonlite::fromJSON(doc)
trainStations = x$results$geometry$location
#head(trainStations)
trainStations$type = "TrainStations"

# malls search
url = paste0("https://maps.googleapis.com/maps/api/place/radarsearch/json?&query=malls+in+banglore&types=shopping_mall&location=12.9716,77.5946&radius=50000&key=",key)
doc <- getURL(url)
x <- jsonlite::fromJSON(doc)
malls = x$results$geometry$location
#head(malls)
malls$type = "Mall"

#####   Processing    #######################

data = rbind(busStations,trainStations, malls)
dim(data)

sample = data
coordinates(sample) <-~ lng +lat # Create cordinates
proj4string(sample) = CRS('+proj=longlat +datum=WGS84') # Add Projections

m<-mcGoogleMaps(sample,zcol = "type", mapTypeId='ROADMAP') # Plot on Google maps


# Get the coordinates
p2 = data[,1:2]

# calculate distances
dist_mat = matrix(0,nrow(p2),nrow(p2))

for (i in 1:nrow(p2)){
  for (j in 1:nrow(p2)){
    dist_mat[i,j] = distCosine(p2[i,],p2[j,], r=6378173)/1000    
  }
}

# Create clusters based in distances
fit <- hclust(as.dist(dist_mat), method="ward")
plot(fit) # display dendogram

groups <- cutree(fit, k=10) # cut tree into 18 clusters
# draw dendogram with red borders around the 18 clusters
rect.hclust(fit, k=10, border="red") 

sample$group = groups # Assign cluster groups

# Plot stores with clustor as label
m <- mcGoogleMaps(sample, mapTypeId='ROADMAP', zcol="group")

Google Map

Dendrogram (grouped into 10 clusters)

Cluster Histogram:

Interpretation

• We found 466 locations clustered in to 10 segments

• After plotting the Google map and visually examining the results, we eliminated the clusters 6, 7, 9 and 10 as the areas corresponding to these clusters were not matching the customer profile

• After eliminating these 4 clusters, it still leaves us with 352 locations from which we need to further narrow down to 4 locations.

• This task has been achieved by setting up an optimization problem discussed in next slide.

Optimization Problem

Results

Assumption:

1. Hypothetical viewership distrubution based on the location
2. Variable cost of the billboard based on the location
3. Monthly ad budget is 10000 INR

---

Note: Because of the arbitrarily assignment of viewership numbers, the final selected 4 might not be the optimal locations. This requires more comprehensive method of collecting the accurate viewership data!

---

We have used Analytic Solver Platform to set up the optimization problem since there were more 200 decision variables.Based on the objective to maximize the viewership while maintaining the cost within the budget, solver provided the below solution:

R Code, to plot selected 4 places:

row1 <- as.numeric(c(12.9370699, 77.6266046))
row2 <-  as.numeric(c(12.9258836, 77.675326))
row3 <-  as.numeric(c(12.9203227, 77.6444846))
row4 <-  as.numeric(c(13.0112861, 77.6630975))

data1 = rbind(row1, row2, row3, row4)
data1 <- data.frame(data1)
colnames(data1) <- c("lat","lng")


sample1 = data1
coordinates(sample1) <-~ lng +lat # Create cordinates
proj4string(sample1) = CRS('+proj=longlat +datum=WGS84') # Add Projections

m<-mcGoogleMaps(sample1,zcol = "type", mapTypeId='ROADMAP') # Plot on Google maps

Below picture depicts the 4 selected areas after plotting them on Google MAP:

---

Synergy Team Members:

* Siva Gangadhar G [71610080]
* Shree Sudhha [71610079]
* Sravya Chunduri [71610110]
* Srikanth Vidapanakal [71610084]

---