Panel_US Prediction Analysis (scale:Quadratic [1-100])

This report analyses the demographic prediction results of Boosted Decision Tree Model trained on the Panel_US data after conversion of ratings on quadratic scale [1-100].

library(ggplot2)
library(reshape)
### load data
predictions <- read.csv("BDT_US_Quadratic.csv")
predictions$RescaledScore <- as.numeric(predictions$RescaledScore)

### split the demo column into two based on spaces
predictions = transform(predictions, new = colsplit(predictions$Demo, split = " ", names = c('Gender', 'Age')))
predictions$title_id <- NULL
predictions$Demo <- NULL
predictions <- setNames(predictions, c("title","score","gender", "age"))
predictions <- predictions[c('title','gender','age','score')]
names(predictions)
## [1] "title"  "gender" "age"    "score"

Top 20 rows of the result set:

## Print top 20 rows
head(predictions,12)
##       title gender     age    score
## 1  Suspects   Male   [65+] 63.81440
## 2  Suspects   Male [45-44] 62.25505
## 3  Suspects   Male [18-24] 70.24672
## 4  Suspects Female [55-64] 66.60513
## 5  Suspects   Male [25-34] 66.60513
## 6  Suspects   Male [35-44] 66.60513
## 7  Suspects Female [25-34] 66.60513
## 8  Suspects   Male [55-64] 57.72056
## 9  Suspects Female [45-44] 73.96347
## 10 Suspects Female [18-24] 66.60513
## 11 Suspects Female   [65+] 68.43636
## 12 Suspects Female [35-44] 66.60513

A summary of the prediction set:

## summarize the data
summary(predictions)
##                          title          gender          age      
##  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                               
##      score       
##  Min.   :  1.00  
##  1st Qu.: 51.77  
##  Median : 57.49  
##  Mean   : 56.42  
##  3rd Qu.: 62.18  
##  Max.   :100.00  
## 

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$score, col=4)

plot of chunk unnamed-chunk-5

The density of the Predicted Scores:

## Check density of scores
d <- density(predictions$score)
plot(d, col=4)

plot of chunk unnamed-chunk-6

Distributional behaviour of predicted scores:

## Check density of scores
boxplot(predictions$score, col=4)

plot of chunk unnamed-chunk-7

Mean Score of each demo bracket:

## get the mean demo scores for the result set
demo <- aggregate(predictions$score, by = predictions[c('gender','age')], mean)

## Sort the data by Gender Values
demo[order(demo$gender),]
##    gender     age        x
## 1  Female [18-24] 56.67109
## 3  Female [25-34] 56.67109
## 5  Female [35-44] 56.67109
## 7  Female [45-44] 63.67327
## 9  Female [55-64] 56.67109
## 11 Female   [65+] 58.41254
## 2    Male [18-24] 60.13491
## 4    Male [25-34] 56.67109
## 6    Male [35-44] 56.67109
## 8    Male [45-44] 52.53745
## 10   Male [55-64] 48.23356
## 12   Male   [65+] 54.01868

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$score, by = predictions[c('gender','age')], sum)

# generate percentages and draw the pie chart
slices <- as.integer(aggregation.age$x)
lbls <- aggregation.age$age
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")

plot of chunk unnamed-chunk-9

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))+geom_bar(width = 1, stat = "identity")
bp

plot of chunk unnamed-chunk-10

Pie Chart indicating Gender by the sum of Scores:

### aggregate the gender by the sum of scores
aggregation.gender <- aggregate(predictions$score, 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")

plot of chunk unnamed-chunk-11

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

plot of chunk unnamed-chunk-12

Heat map of Demo by predictions scores on all titles:

library(reshape2)
## 
## Attaching package: 'reshape2'
## The following objects are masked from 'package:reshape':
## 
##     colsplit, melt, recast
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$age <- factor(predictions$age, 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),]

### dcast the data to convert into matrix of Title, Demo

demo.data <- dcast(predictions, title~gender+age)
## Using score 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("yellow", "green", "red"))

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(15,80), key= T, key.xlab="Predeiction_Score") 

plot of chunk unnamed-chunk-13

Titles By Gender Split:

aggregation.gender <- aggregate(predictions$score, 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()

plot of chunk unnamed-chunk-14

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$score))
## 
##     1 
## 18288