The music industry has a well-developed market with a global revenue valued at $16.5 billion in 2012. The recording industry is highly competitive and is dominated by three big production companies which make up nearly 82% of the total annual album sales.
Artists are at the core of the music industry and record labels provide them with the necessary resources to sell their music on a large scale. A record label incurs numerous costs (studio recording, marketing, distribution, and touring) in exchange for a percentage of the profits from album sales, singles and concert tickets.
Unfortunately, the success of an artist's release is highly uncertain: a single may be extremely popular, resulting in widespread radio play and digital downloads, while another single may turn out quite unpopular, and therefore unprofitable.
Knowing the competitive nature of the recording industry, record labels face the fundamental decision problem of which musical releases to support to maximize their financial success.
Analytics is used to predict whether a song will reach a spot in the Top 10 of the Billboard Hot 100 Chart. The dataset consists of all songs which made it to the Top 10 of the Billboard Hot 100 Chart from 1990-2010 plus a sample of additional songs that didn't make the Top 10. This data comes from three sources: Wikipedia, Billboard.com, and EchoNest.
The variables included in the dataset either describe the artist or the song, or they are associated with the following song attributes: time signature, loudness, key, pitch, tempo, and timbre.
Here's a detailed description of the variables:
year = the year the song was released
songtitle = the title of the song
artistname = the name of the artist of the song
songID and artistID = identifying variables for the song and artist
timesignature and timesignature_confidence = a variable estimating the time signature of the song, and the confidence in the estimate
loudness = a continuous variable indicating the average amplitude of the audio in decibels
tempo and tempo_confidence = a variable indicating the estimated beats per minute of the song, and the confidence in the estimate
key and key_confidence = a variable with twelve levels indicating the estimated key of the song (C, C#, . . ., B), and the confidence in the estimate
energy = a variable that represents the overall acoustic energy of the song, using a mix of features such as loudness
pitch = a continuous variable that indicates the pitch of the song
timbre_0_min, timbre_0_max, timbre_1_min, timbre_1_max, . . . , timbre_11_min, and timbre_11_max = variables that indicate the minimum/maximum values over all segments for each of the twelve values in the timbre vector (resulting in 24 continuous variables)
Top10 = a binary variable indicating whether or not the song made it to the Top 10 of the Billboard Hot 100 Chart (1 if it was in the top 10, and 0 if it was not)
# Read the data
songs <- read.csv("songs.csv")
# Subset of the data set (year=2010)
songs.2010 = subset(songs, year == 2010)
# Songs in 2010
nrow(songs.2010)
## [1] 373
# Number of songs with artist name is Michael Jackson
MichaelJackson = subset(songs, artistname == "Michael Jackson")
nrow(MichaelJackson)
## [1] 18
# Songs made to top10 by Michael Jackson
MichaelJackson.10 = subset(songs, artistname == "Michael Jackson" & Top10 ==
1)
levels(factor(MichaelJackson.10$songtitle))
## [1] "Black or White" "In The Closet" "Remember the Time"
## [4] "You Are Not Alone" "You Rock My World"
# timesignature variable
table(songs$timesignature)
##
## 0 1 3 4 5 7
## 10 143 503 6787 112 19
# Song with highest tempo
levels(factor(songs$songtitle[which.max(songs$tempo)]))
## [1] "Wanna Be Startin' Somethin'"
# Split the data into training and test sets
SongsTrain <- subset(songs, year <= 2009)
SongsTest <- subset(songs, year == 2010)
# define a vector of variables that we won't use in our model.
nonvars = c("year", "songtitle", "artistname", "songID", "artistID")
# Remove these variables from the training and test sets
SongsTrain = SongsTrain[, !(names(SongsTrain) %in% nonvars)]
SongsTest = SongsTest[, !(names(SongsTest) %in% nonvars)]
# Build the model
Model1 <- glm(Top10 ~ ., data = SongsTrain, family = "binomial")
summary(Model1)
##
## Call:
## glm(formula = Top10 ~ ., family = "binomial", data = SongsTrain)
##
## Deviance Residuals:
## Min 1Q Median 3Q Max
## -1.922 -0.540 -0.346 -0.184 3.077
##
## Coefficients:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) 1.47e+01 1.81e+00 8.14 4.0e-16 ***
## timesignature 1.26e-01 8.67e-02 1.46 0.14505
## timesignature_confidence 7.45e-01 1.95e-01 3.81 0.00014 ***
## loudness 3.00e-01 2.92e-02 10.28 < 2e-16 ***
## tempo 3.63e-04 1.69e-03 0.21 0.82989
## tempo_confidence 4.73e-01 1.42e-01 3.33 0.00087 ***
## key 1.59e-02 1.04e-02 1.53 0.12635
## key_confidence 3.09e-01 1.41e-01 2.19 0.02876 *
## energy -1.50e+00 3.10e-01 -4.85 1.3e-06 ***
## pitch -4.49e+01 6.83e+00 -6.57 5.0e-11 ***
## timbre_0_min 2.32e-02 4.26e-03 5.44 5.3e-08 ***
## timbre_0_max -3.31e-01 2.57e-02 -12.88 < 2e-16 ***
## timbre_1_min 5.88e-03 7.80e-04 7.54 4.6e-14 ***
## timbre_1_max -2.45e-04 7.15e-04 -0.34 0.73209
## timbre_2_min -2.13e-03 1.13e-03 -1.89 0.05884 .
## timbre_2_max 6.59e-04 9.07e-04 0.73 0.46757
## timbre_3_min 6.92e-04 5.98e-04 1.16 0.24758
## timbre_3_max -2.97e-03 5.81e-04 -5.10 3.3e-07 ***
## timbre_4_min 1.04e-02 1.98e-03 5.24 1.6e-07 ***
## timbre_4_max 6.11e-03 1.55e-03 3.94 8.1e-05 ***
## timbre_5_min -5.60e-03 1.28e-03 -4.38 1.2e-05 ***
## timbre_5_max 7.74e-05 7.94e-04 0.10 0.92234
## timbre_6_min -1.69e-02 2.26e-03 -7.45 9.7e-14 ***
## timbre_6_max 3.67e-03 2.19e-03 1.68 0.09388 .
## timbre_7_min -4.55e-03 1.78e-03 -2.55 0.01066 *
## timbre_7_max -3.77e-03 1.83e-03 -2.06 0.03941 *
## timbre_8_min 3.91e-03 2.85e-03 1.37 0.17012
## timbre_8_max 4.01e-03 3.00e-03 1.34 0.18162
## timbre_9_min 1.37e-03 3.00e-03 0.46 0.64836
## timbre_9_max 1.60e-03 2.43e-03 0.66 0.51019
## timbre_10_min 4.13e-03 1.84e-03 2.24 0.02485 *
## timbre_10_max 5.83e-03 1.77e-03 3.29 0.00099 ***
## timbre_11_min -2.63e-02 3.69e-03 -7.11 1.2e-12 ***
## timbre_11_max 1.97e-02 3.39e-03 5.81 6.2e-09 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## (Dispersion parameter for binomial family taken to be 1)
##
## Null deviance: 6017.5 on 7200 degrees of freedom
## Residual deviance: 4759.2 on 7167 degrees of freedom
## AIC: 4827
##
## Number of Fisher Scoring iterations: 6
# Correlation between loudness and energy
cor(songs$loudness, songs$energy) # loudness and energy are highly correlated
## [1] 0.742
# Create a model without the variable loudness
Model2 = glm(Top10 ~ . - loudness, data = SongsTrain, family = binomial)
summary(Model2)
##
## Call:
## glm(formula = Top10 ~ . - loudness, family = binomial, data = SongsTrain)
##
## Deviance Residuals:
## Min 1Q Median 3Q Max
## -2.098 -0.561 -0.360 -0.190 3.311
##
## Coefficients:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) -2.24e+00 7.46e-01 -3.00 0.00269 **
## timesignature 1.62e-01 8.73e-02 1.86 0.06287 .
## timesignature_confidence 6.88e-01 1.92e-01 3.58 0.00035 ***
## tempo 5.52e-04 1.67e-03 0.33 0.74023
## tempo_confidence 5.50e-01 1.41e-01 3.91 9.4e-05 ***
## key 1.74e-02 1.03e-02 1.70 0.08974 .
## key_confidence 2.95e-01 1.39e-01 2.12 0.03416 *
## energy 1.81e-01 2.61e-01 0.70 0.48699
## pitch -5.15e+01 6.86e+00 -7.51 5.9e-14 ***
## timbre_0_min 2.48e-02 4.24e-03 5.85 5.0e-09 ***
## timbre_0_max -1.01e-01 1.18e-02 -8.55 < 2e-16 ***
## timbre_1_min 7.14e-03 7.71e-04 9.27 < 2e-16 ***
## timbre_1_max -7.83e-04 7.06e-04 -1.11 0.26765
## timbre_2_min -1.58e-03 1.11e-03 -1.42 0.15453
## timbre_2_max 3.89e-04 8.96e-04 0.43 0.66443
## timbre_3_min 6.50e-04 5.95e-04 1.09 0.27452
## timbre_3_max -2.46e-03 5.67e-04 -4.34 1.4e-05 ***
## timbre_4_min 9.11e-03 1.95e-03 4.67 3.0e-06 ***
## timbre_4_max 6.31e-03 1.53e-03 4.12 3.9e-05 ***
## timbre_5_min -5.64e-03 1.26e-03 -4.50 7.0e-06 ***
## timbre_5_max 6.94e-04 7.81e-04 0.89 0.37426
## timbre_6_min -1.61e-02 2.24e-03 -7.21 5.5e-13 ***
## timbre_6_max 3.81e-03 2.16e-03 1.77 0.07698 .
## timbre_7_min -5.10e-03 1.76e-03 -2.91 0.00364 **
## timbre_7_max -3.16e-03 1.81e-03 -1.74 0.08109 .
## timbre_8_min 4.49e-03 2.81e-03 1.60 0.11025
## timbre_8_max 6.42e-03 2.95e-03 2.18 0.02950 *
## timbre_9_min -4.28e-04 2.96e-03 -0.14 0.88479
## timbre_9_max 3.52e-03 2.38e-03 1.48 0.13802
## timbre_10_min 2.99e-03 1.80e-03 1.66 0.09700 .
## timbre_10_max 7.37e-03 1.73e-03 4.25 2.1e-05 ***
## timbre_11_min -2.84e-02 3.63e-03 -7.82 5.5e-15 ***
## timbre_11_max 1.83e-02 3.34e-03 5.48 4.3e-08 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## (Dispersion parameter for binomial family taken to be 1)
##
## Null deviance: 6017.5 on 7200 degrees of freedom
## Residual deviance: 4871.8 on 7168 degrees of freedom
## AIC: 4938
##
## Number of Fisher Scoring iterations: 6
# Create a model without the variable energy
Model3 = glm(Top10 ~ . - energy, data = SongsTrain, family = binomial)
summary(Model3)
##
## Call:
## glm(formula = Top10 ~ . - energy, family = binomial, data = SongsTrain)
##
## Deviance Residuals:
## Min 1Q Median 3Q Max
## -1.918 -0.542 -0.348 -0.187 3.417
##
## Coefficients:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) 1.20e+01 1.71e+00 6.98 3.0e-12 ***
## timesignature 1.15e-01 8.73e-02 1.32 0.18718
## timesignature_confidence 7.14e-01 1.95e-01 3.67 0.00024 ***
## loudness 2.31e-01 2.53e-02 9.12 < 2e-16 ***
## tempo -6.46e-04 1.67e-03 -0.39 0.69811
## tempo_confidence 3.84e-01 1.40e-01 2.75 0.00602 **
## key 1.65e-02 1.04e-02 1.59 0.11106
## key_confidence 3.39e-01 1.41e-01 2.41 0.01598 *
## pitch -5.33e+01 6.73e+00 -7.91 2.5e-15 ***
## timbre_0_min 2.20e-02 4.24e-03 5.20 2.0e-07 ***
## timbre_0_max -3.10e-01 2.54e-02 -12.24 < 2e-16 ***
## timbre_1_min 5.42e-03 7.64e-04 7.09 1.4e-12 ***
## timbre_1_max -5.12e-04 7.11e-04 -0.72 0.47193
## timbre_2_min -2.25e-03 1.12e-03 -2.01 0.04419 *
## timbre_2_max 4.12e-04 9.02e-04 0.46 0.64791
## timbre_3_min 3.18e-04 5.87e-04 0.54 0.58808
## timbre_3_max -2.96e-03 5.76e-04 -5.15 2.6e-07 ***
## timbre_4_min 1.10e-02 1.98e-03 5.58 2.3e-08 ***
## timbre_4_max 6.47e-03 1.54e-03 4.20 2.7e-05 ***
## timbre_5_min -5.13e-03 1.27e-03 -4.05 5.2e-05 ***
## timbre_5_max 2.98e-04 7.86e-04 0.38 0.70453
## timbre_6_min -1.78e-02 2.25e-03 -7.94 1.9e-15 ***
## timbre_6_max 3.45e-03 2.18e-03 1.58 0.11420
## timbre_7_min -5.13e-03 1.77e-03 -2.90 0.00373 **
## timbre_7_max -3.39e-03 1.82e-03 -1.86 0.06221 .
## timbre_8_min 3.69e-03 2.83e-03 1.30 0.19323
## timbre_8_max 4.66e-03 2.99e-03 1.56 0.11902
## timbre_9_min -9.32e-05 2.96e-03 -0.03 0.97486
## timbre_9_max 1.34e-03 2.42e-03 0.55 0.57990
## timbre_10_min 4.05e-03 1.83e-03 2.22 0.02664 *
## timbre_10_max 5.79e-03 1.76e-03 3.29 0.00099 ***
## timbre_11_min -2.64e-02 3.68e-03 -7.16 8.0e-13 ***
## timbre_11_max 1.98e-02 3.36e-03 5.90 3.7e-09 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## (Dispersion parameter for binomial family taken to be 1)
##
## Null deviance: 6017.5 on 7200 degrees of freedom
## Residual deviance: 4782.7 on 7168 degrees of freedom
## AIC: 4849
##
## Number of Fisher Scoring iterations: 6
# Make predictions on the test set
predictTest <- predict(Model3, newdata = SongsTest, type = "response")
# Confusion Matrix with threshold of 0.45
confMat <- table(SongsTest$Top10, predictTest >= 0.45)
# Accuracy
(confMat[1, 1] + confMat[2, 2])/sum(confMat)
## [1] 0.8794
# Baseline model. Pick the most frequent outcome (a song is not a Top 10
# hit) for all songs
baseline <- table(SongsTest$Top10)
# Accuracy
baseline[1]/(baseline[1] + baseline[2])
## 0
## 0.8418
It seems that Model 3 gives us a small improvement over the baseline model. Let's view the two models from an investment perspective. A production company is interested in investing in songs that are highly likely to make it to the Top 10. The company's objective is to minimize its risk of financial losses attributed to investing in songs that end up unpopular.
A competitive edge can therefore be achieved if we can provide the production company a list of songs that are highly likely to end up in the Top 10. We note that the baseline model does not prove useful, as it simply does not label any song as a hit. Let us see what our model has to offer.
# Sensitivity and Specificity
confMat[2, 2]/as.numeric(rowSums(confMat)[2])
## [1] 0.322
confMat[1, 1]/as.numeric(rowSums(confMat)[1])
## [1] 0.9841
Model 3 has a very high specificity, meaning that it favors specificity over sensitivity. While Model 3 only captures less than half of the Top 10 songs, it still can offer a competitive edge, since it is very conservative in its predictions.