This report analyses the demographic prediction results of Boosted Decision Tree Model trained on the Panel_US data after conversion of ratings on linear scale [1-100].
## from the knite WD: C:\Users\Admin\Documents\R\Demo_Data_Prep\
predictions <- read.csv("BDT_US_ScaleConverted.csv")
colnames(predictions)
## [1] "Title" "Gender" "Age.Bracket" "Scored.Labels"
predictions$Scored.Labels <- as.numeric(predictions$Scored.Labels)
class(predictions$Scored.Labels)
## [1] "numeric"
library(ggplot2)
Top 20 rows of the result set:
## Print top 20 rows
head(predictions,12)
## Title Gender Age.Bracket Scored.Labels
## 1 Suspects Male [35-44] 67.56518
## 2 Suspects Male [55-64] 60.12343
## 3 Suspects Female [55-64] 67.56518
## 4 Suspects Female [35-44] 67.56518
## 5 Suspects Male [18-24] 70.52461
## 6 Suspects Female [18-24] 67.56518
## 7 Suspects Male [25-34] 67.56518
## 8 Suspects Female [25-34] 67.56518
## 9 Suspects Male [65+] 65.26256
## 10 Suspects Female [65+] 69.05964
## 11 Suspects Male [45-44] 63.96233
## 12 Suspects Female [45-44] 73.49471
A summary of the prediction set:
## summarize the data
summary(predictions)
## Title Gender Age.Bracket
## 10 Things I Hate About You : 12 Female:9144 [18-24]:3048
## 100 Deeds For Eddie Mcdowd : 12 Male :9144 [25-34]:3048
## 14 Diaries Of The Great War: 12 [35-44]:3048
## 1600 Penn : 12 [45-44]:3048
## 17 Kids And Counting : 12 [55-64]:3048
## 2 Broke Girls : 12 [65+] :3048
## (Other) :18216
## Scored.Labels
## Min. : 1.01
## 1st Qu.:54.95
## Median :59.92
## Mean :58.79
## 3rd Qu.:63.90
## Max. :93.06
##
Count of Unique Titles in the Test Set:
## count of unique titles in the result set
length(unique(predictions$Title))
## [1] 1524
Histograpm of Predicted Scores:
hist(predictions$Scored.Labels, col=4)
The density of the Predicted Scores:
## Check density of scores
d <- density(predictions$Scored.Labels, col=4)
## Warning: In density.default(predictions$Scored.Labels, col = 4) :
## extra argument 'col' will be disregarded
plot(d)
Distributional behaviour of predicted scores:
## Check density of scores
boxplot(predictions$Scored.Labels, col=4)
Mean Score of each demo bracket:
## get the mean demo scores for the result set
demo <- aggregate(predictions$Scored.Labels, by = predictions[c('Gender','Age.Bracket')], mean)
## Sort the data by Gender Values
demo[order(demo$Gender),]
## Gender Age.Bracket x
## 1 Female [18-24] 59.03567
## 3 Female [25-34] 59.03567
## 5 Female [35-44] 59.03567
## 7 Female [45-44] 64.96521
## 9 Female [55-64] 59.03567
## 11 Female [65+] 60.53011
## 2 Male [18-24] 61.99509
## 4 Male [25-34] 59.03567
## 6 Male [35-44] 59.03567
## 8 Male [45-44] 55.43282
## 10 Male [55-64] 51.59392
## 12 Male [65+] 56.73304
Pie Chart Age brackets by the sum of Scores:
### Create a pie chart of the demo ages by percentage based on prediction scores
# create an aggregate view of age brackets by sum of scores
aggregation.age <- aggregate(predictions$Scored.Labels, by = predictions['Age.Bracket'], sum)
# generate percentages and draw the pie chart
slices <- as.integer(aggregation.age$x)
lbls <- aggregation.age$Age.Bracket
pct <- round(slices/sum(slices)*100)
lbls <- paste(lbls, pct) # add percents to labels
lbls <- paste(lbls,"%",sep="") # ad % to labels
pie(slices,labels = lbls, col=rainbow(length(lbls)),main="Age Brackets By Sum of Scores")
Bar chart indicating the ranked Age brackets by the sum of scores:
### aggregate the age brackets by the sum of scores
aggregation.age <- aggregation.age[order(-aggregation.age$x),]
# draw Bar chart
bp<- ggplot(aggregation.age, aes(x="", y=x, fill =Age.Bracket))+geom_bar(width = 1, stat = "identity")
bp
Pie Chart indicating Gender by the sum of Scores:
### aggregate the gender by the sum of scores
aggregation.gender <- aggregate(predictions$Scored.Labels, by = predictions['Gender'], sum)
### Create a pie chart of the demo ages by percentage based on prediction scores
slices <- as.integer(aggregation.gender$x)
lbls <- aggregation.gender$Gender
pct <- round(slices/sum(slices)*100)
lbls <- paste(lbls, pct) # add percents to labels
lbls <- paste(lbls,"%",sep="") # ad % to labels
# draw pie chart
pie(slices,labels = lbls, col = c("violetred1", "blue"),main="Pie Chart of Age Brackets By Sum of Scores")
Bar chart indicating the gender by the sum of scores:
# aggregate gender data by sum of scores
aggregation.gender <- aggregation.gender[order(-aggregation.gender$x),]
# rename new column
names(aggregation.gender)[names(aggregation.gender)=="x"] <- "Score_Sum"
# draw Bar chart
bp<- ggplot(aggregation.gender, aes(x=Gender, y=Score_Sum, fill =Gender))+geom_bar(width = 1, stat = "identity")
bp
Heat map of Demo by predictions scores on all titles:
library(reshape2)
library(gplots)
##
## Attaching package: 'gplots'
## The following object is masked from 'package:stats':
##
## lowess
# sett the sorting order
predictions$Gender <- factor(predictions$Gender, levels = c("Female", "Male"))
predictions$Age.Bracket <- factor(predictions$Age.Bracket, levels = c("[18-24]","[25-34]","[35-44]","[45-44]","[55-64]","[65+]"))
#sort the data based on gender and age for each title
predictions <- predictions[order(predictions$Title, predictions$Gender,predictions$Age.Bracket),]
### dcast the data to convert into matrix of Title, Demo
demo.data <- dcast(predictions, Title~Gender+Age.Bracket)
## Using Scored.Labels as value column: use value.var to override.
# set titles as row names and get rid of this column
row.names(demo.data) <- demo.data$Title
dim(demo.data)
## [1] 1524 13
# get the numberic data only
demo.data <- demo.data[,2:13]
# convert to a numeric matrix
demo_matrix <- data.matrix(demo.data)
# create a data subset for visualization
demo.sub <- head(demo_matrix, 30)
# creates a own color palette from red to green
my_palette <- colorRampPalette(c("green", "red", "yellow"))
heatmap.2(demo.sub, scale = "none", col=my_palette, trace="none", dendrogram=c("none"), symm=F,symkey=F,symbreaks=T, cexCol=0.7, cexRow=0.6,density.info="histogram",breaks = seq(1, 100), key= T, key.xlab="Predeiction_Score")
Titles By Gender Split:
aggregation.gender <- aggregate(predictions$Scored.Labels, by = predictions[c('Title','Gender')], sum)
names(aggregation.gender)[names(aggregation.gender)=="x"] <- "gender.score"
# order the data by title, gender and score
aggregation.gender <- aggregation.gender[order(aggregation.gender$Title, aggregation.gender$Gender,aggregation.gender$gender.score),]
#plot the stacked bar chart
ggplot(data = head(aggregation.gender,100), aes(x = Title, y =gender.score, fill = Gender)) + geom_bar(stat="identity") + coord_flip()
Count of Female Titles:
### dcast the data to convert into matrix of Title, Male and Female
gender.dcast <- dcast(aggregation.gender, Title~Gender)
## Using gender.score as value column: use value.var to override.
# count of Female shows
length(unique(gender.dcast$Title[gender.dcast$Male < gender.dcast$Female]))
## [1] 1524
Count of Male Titles:
# count of male shows
length(unique(gender.dcast$Title[gender.dcast$Male > gender.dcast$Female]))
## [1] 0
Count of Negative predictions:
## Print count of negative and positive predictions on all demo
table(sign(predictions$Scored.Labels))
##
## 1
## 18288