Cek Missing Value

##                        X            playlist_name               track.name 
##                        0                        0                        0 
##      playlist_owner_name             danceability                   energy 
##                        0                        0                        0 
##                 loudness              speechiness             acousticness 
##                        0                        0                        0 
##         instrumentalness                 liveness                  valence 
##                        0                        0                        0 
##                    tempo         track.popularity         track.album.name 
##                        0                        0                        0 
## track.album.release_date        track.duration_ms 
##                        0                        0

Tidak terdapat missing value pada data kita.

Berikutnya terlebih dahulu kita periksa summary dari data frame my_playlist. Sebagai tambahan informasi, berikut penjelasan terkait apa saja yang terdapat pada audio feature di data kita berdasarkan Spotify API Docs sebagaimana dalam membantu kita mengerjakan project ini:

• acousticness: A confidence measure from 0.0 to 1.0 of whether the track is acoustic. 1.0 represents high confidence the track is acoustic.

• danceability: Danceability describes how suitable a track is for dancing based on a combination of musical elements including tempo, rhythm stability, beat strength, and overall regularity. A value of 0.0 is least danceable and 1.0 is most danceable.

• energy: Energy is a measure from 0.0 to 1.0 and represents a perceptual measure of intensity and activity. Typically, energetic tracks feel fast, loud, and noisy. For example, death metal has high energy, while a Bach prelude scores low on the scale. Perceptual features contributing to this attribute include dynamic range, perceived loudness, timbre, onset rate, and general entropy.

• instrumentalness: Predicts whether a track contains no vocals. “Ooh” and “aah” sounds are treated as instrumental in this context. Rap or spoken word tracks are clearly “vocal”. The closer the instrumentalness value is to 1.0, the greater likelihood the track contains no vocal content. Values above 0.5 are intended to represent instrumental tracks, but confidence is higher as the value approaches 1.0.

• liveness: Detects the presence of an audience in the recording. Higher liveness values represent an increased probability that the track was performed live. A value above 0.8 provides strong likelihood that the track is live.

• loudness: The overall loudness of a track in decibels (dB). Loudness values are averaged across the entire track and are useful for comparing relative loudness of tracks. Loudness is the quality of a sound that is the primary psychological correlate of physical strength (amplitude). Values typically range between -60 and 0 db.

• speechiness: Speechiness detects the presence of spoken words in a track. The more exclusively speech-like the recording (e.g. talk show, audio book, poetry), the closer to 1.0 the attribute value. Values above 0.66 describe tracks that are probably made entirely of spoken words. Values between 0.33 and 0.66 describe tracks that may contain both music and speech, either in sections or layered, including such cases as rap music. Values below 0.33 most likely represent music and other non-speech-like tracks.

Cek summary my_playlist

##        X         playlist_name       track.name        playlist_owner_name
##  Min.   :  1.0   Length:924         Length:924         Length:924         
##  1st Qu.:231.8   Class :character   Class :character   Class :character   
##  Median :462.5   Mode  :character   Mode  :character   Mode  :character   
##  Mean   :462.5                                                            
##  3rd Qu.:693.2                                                            
##  Max.   :924.0                                                            
##   danceability        energy          loudness        speechiness     
##  Min.   :0.2360   Min.   :0.1930   Min.   :-18.118   Min.   :0.02400  
##  1st Qu.:0.5630   1st Qu.:0.4850   1st Qu.: -7.810   1st Qu.:0.03950  
##  Median :0.6630   Median :0.6280   Median : -6.309   Median :0.05270  
##  Mean   :0.6376   Mean   :0.6067   Mean   : -6.559   Mean   :0.08095  
##  3rd Qu.:0.7210   3rd Qu.:0.7390   3rd Qu.: -4.965   3rd Qu.:0.08360  
##  Max.   :0.9130   Max.   :0.9600   Max.   : -1.856   Max.   :0.61100  
##   acousticness     instrumentalness      liveness          valence      
##  Min.   :0.00125   Min.   :0.000000   Min.   :0.01900   Min.   :0.0389  
##  1st Qu.:0.04620   1st Qu.:0.000000   1st Qu.:0.09527   1st Qu.:0.3098  
##  Median :0.19600   Median :0.000000   Median :0.11700   Median :0.4595  
##  Mean   :0.27244   Mean   :0.006044   Mean   :0.16561   Mean   :0.4762  
##  3rd Qu.:0.43550   3rd Qu.:0.000054   3rd Qu.:0.18425   3rd Qu.:0.6240  
##  Max.   :0.95500   Max.   :0.833000   Max.   :0.89200   Max.   :0.9560  
##      tempo        track.popularity track.album.name   track.album.release_date
##  Min.   : 45.78   Min.   :  0.00   Length:924         Length:924              
##  1st Qu.: 94.72   1st Qu.: 64.00   Class :character   Class :character        
##  Median :115.46   Median : 73.00   Mode  :character   Mode  :character        
##  Mean   :117.27   Mean   : 69.16                                              
##  3rd Qu.:134.06   3rd Qu.: 79.00                                              
##  Max.   :199.89   Max.   :100.00                                              
##  track.duration_ms
##  Min.   :108613   
##  1st Qu.:173929   
##  Median :187488   
##  Mean   :194329   
##  3rd Qu.:213837   
##  Max.   :588139

Dari hasil summary() terdapat beberapa insights:

1. Terdapat kolom yang kita tidak perlukan yaitu kolom X

2. Perlu kita ubah tipe data dari kolom berikut yaitu playlist_name, playlist_owner_name, dan track.album.release_date

3. Perlu kita rapihkan format dari kolom track.duration_ms menjadi bentuk %M:%S

4. Berdasarkan Spotify API Docs untuk tiap audio feature di setiap track kebanyakan memiliki rentang nilai 0 sampai 1. Namun dari hasil summary bahwa kolom danceability, energy, loudness, speechiness, acousticness, instrumentalness, liveness, valence perlu kita normalisasikan agar memiliki rentang nilai 0 sampai 1

Kita rapihkan data kita

# wrangling
playlist.clean <- my_playlist %>%
  select(-X) %>% 
  mutate_at(.vars = c("playlist_name","playlist_owner_name"), .funs = as.factor) %>%
  mutate(track.album.release_date = as.Date(track.album.release_date)) %>% 
  mutate(track.duration_ms = format(
    as.POSIXct(Sys.Date(),tz = "GMT")+track.duration_ms/1000,"%M:%S"))

# rename
playlist.clean <- playlist.clean %>% 
  rename(track.duration = track.duration_ms)

Popular Track

Berikutnya coba kita lihat lagu apa saja yang paling saya dengar berdasarkan nilai popularitas nya

Dari hasil grafik diketahui bahwa lagu As It Was oleh Harry Styles adalah lagu populer yang sering saya dengarkan. Selanjutnya kita akan lihat lagu-lagu keluaran tahun apa yang sering saya dengar

Songs by Year

Dari hasil barplot diketahui bahwa lagu yang sering saya dengarkan adalah lagu keluaran di tahun 2022. Selanjutnya kita lakukan analisa insight apa yang kita dapat dari hubungan antar parameter-parameter dari audio feature

Heatmap correlation audio feature

Melalui plot heatmap dapat kita temukan beberapa insights:

1. energy dan loudness berkorelasi kuat positif (0,8)

2. danceability dan valence berkorelasi positif (0,5)

3. energy dan loudness berkorelasi positif terhadap valence (0,3)

Berikutnya akan kita lihat hubungan dari insights yang kita temukan ke dalam bentuk visual scatterplot

Dari hasil scatterplot ditemukan bahwa seluruh variabel berkorelasi linear dengan korelasi kuat positif terjadi pada energy dengan loudness

Berikutnya kita lihat sebaran audio_feature dari masing-masing playlist melalui radar chart. Namun kita akan cek terlebih dahulu outlier pada data

Cek outlier

Removing outlier

outlier <- c(271,286,316,317,665,666,680,709,747,748)

playlist_normal <- playlist.clean[-outlier,]

Setelah membuang outiler, kita lakukan tahap normalisasi dengan metode min-max scaler

# rownames() track

data.playlist <- playlist_normal %>% 
  distinct(track.name, .keep_all = T) %>% 
  as.tibble() %>% 
  column_to_rownames(var = "track.name") %>% 
  select(-c("playlist_name","playlist_owner_name",
            "tempo", "track.popularity", "track.album.name",
            "track.album.release_date","track.duration"))


# Normalize the data (Has 0-1 range)
normalize <- function(x) {
  return((x - min(x)) / (max(x) - min(x)))
}


data.playlist['acousticness'] <- normalize(data.playlist['acousticness'])
data.playlist['danceability'] <- normalize(data.playlist['danceability'])
data.playlist['energy'] <- normalize(data.playlist['energy'])
data.playlist['instrumentalness'] <- normalize(data.playlist['instrumentalness'])
data.playlist['liveness'] <- normalize(data.playlist['liveness'])
data.playlist['loudness'] <- normalize(data.playlist['loudness'])
data.playlist['speechiness'] <- normalize(data.playlist['speechiness'])
data.playlist['valence'] <- normalize(data.playlist['valence'])


data.playlist$avg_feature <- data.playlist %>% 
  rowMeans()

data_radar <- data.playlist %>% 
  arrange(-avg_feature)

# Add the maximum, minimum row value above our dataset
playlist.web <- rbind(1,0 , data_radar) %>% 
  select(-avg_feature) %>% 
  head(15)

Playlist Analysis

Dari plot radar chart ditemukan bahwa audio feature masing-masing playlist sebagai berikut,

• keseluruhan playlist memiliki instrumentalness yang rendah

• playlist My RnB mendominasi pada bagian speechiness dan acousticness

• playlist mood hanya mendominasi bagian acousticness sedangkan playlist Pop Rising hanya mendominasi bagian valence

• playlist Are & Be mendominasi tiga bagian audio feature yaitu energy,loudness, dandanceability sedangkan playlist Workout mendominasi di bagian loudness, energy, dan valence

• playlist Alternative R&B mendominasi dua bagian audio feature yaitu speechiness dan liveness

Tahap berikutnya menentukan nilai k optimum dengan membandingkan antara metode Elbow method dengan Silhouette Method. Hal ini dikarenakan dalam project ini kita akan menggunakan metode clustering dengan algoritma K-Means. Sehingga diperlukan berapa nilai optimum k yang dibutuhkan

Elbow method vs Silhouette Method

Berdasarkan hasil dari grafik elbow method kita akan menggunakan nilai k optimum yaitu 4 dari hasil Elbow Method

Setelah kita mengetahui nilai k yang optimum, berikutnya kita bangun algoritma K-Means

K-Means Cluster

Cluster Profiling

# melakukan profiling dengan summarise data
playlist.profil <- data.playlist %>% 
  group_by(cluster) %>% 
  summarise_all(mean)

playlist.profil %>% 
  pivot_longer(-cluster, names_to = "audio_feature") %>% 
  group_by(audio_feature) %>% 
  summarize(cluster_min_val = which.min(value),
            cluster_max_val = which.max(value))

## # A tibble: 8 x 3
##   audio_feature    cluster_min_val cluster_max_val
##   <chr>                      <int>           <int>
## 1 acousticness                   3               1
## 2 danceability                   1               3
## 3 energy                         1               3
## 4 instrumentalness               3               1
## 5 liveness                       1               4
## 6 loudness                       1               3
## 7 speechiness                    3               2
## 8 valence                        4               3

Melalui hasil profiling cluster kita temukan karakteristik dari masing-masing klaster.

• Cluster 1 (Chill) : High acousticness and instrumentalness; low danceability, energy,danceability, liveness, loudness

## # A tibble: 5 x 1
##   value         
##   <chr>         
## 1 Supermodel    
## 2 Drew Barrymore
## 3 The Weekend   
## 4 Go Gina       
## 5 20 Something

• Cluster 2 (Romantic) : High speechiness

## # A tibble: 5 x 1
##   value                   
##   <chr>                   
## 1 Because Of You          
## 2 So Sick                 
## 3 My Boo                  
## 4 Garden (Say It Like Dat)
## 5 No Love (with SZA)

• Cluster 3 (Energetic) : High danceability, energy, loudness; low acousticness, instrumentalness, speechiness

## # A tibble: 5 x 1
##   value                   
##   <chr>                   
## 1 Kiss Me More (feat. SZA)
## 2 Hit Different           
## 3 All The Stars (with SZA)
## 4 Consideration           
## 5 With You

• Cluster 4 (Sad) : High liveness, low valence

## # A tibble: 5 x 1
##   value                           
##   <chr>                           
## 1 No Air (feat. Chris Brown)      
## 2 No One                          
## 3 Love Galore (feat. Travis Scott)
## 4 Prom                            
## 5 Broken Clocks

KNN

# build knn

knn_mood <- knn(train = mood_train[,-9],
                    test = mood_test[,-9],
                    cl = mood_train$cluster,
                    k = round(sqrt((nrow(mood_train))),2))
# confusion matrix

knn_result <- confusionMatrix(as.factor(knn_mood), 
                as.factor(mood_test_y))

df_knn_result <- as.data.frame(as.table(knn_result)) %>% 
  mutate(
    model = 'KNN'
  )

model_result <- df_knn_result

# result for Accuracy
acc_knn <- as.data.frame(as.table(
  as.matrix(knn_result, 
            what = "overall"))) %>% 
  mutate(
    Model = 'KNN'
  ) %>% 
  filter(
    Var1 == 'Accuracy'
  ) %>% 
  rename(
    Accuracy = Freq
  ) %>% 
  select(Model, Accuracy)

accuracy <- acc_knn

Naive Bayes

# build model naive bayes

naives_mood <- naiveBayes(mood_train %>% select(-cluster),
                          mood_train$cluster)

# predict

naive_pred <- predict(naives_mood, mood_test, type = "class")

# confusion matrix
naive_result <- confusionMatrix(as.factor(naive_pred), 
                                as.factor(mood_test$cluster))

df_naive_result <- as.data.frame(as.table(naive_result)) %>% 
  mutate(
    model = 'NAIVE BAYES'
  )

model_result <- rbind(model_result, df_naive_result)

# result for Accuracy
acc_naive <- as.data.frame(as.table(
  as.matrix(naive_result, 
            what = "overall"))) %>% 
  mutate(
    Model = 'NAIVE BAYES'
  ) %>% 
  filter(
    Var1 == 'Accuracy'
  ) %>% 
  rename(
    Accuracy = Freq
  ) %>% 
  select(Model, Accuracy)

accuracy <- rbind(accuracy, acc_naive)

Random Forest

library(partykit)

set.seed(123)

ctrl <- trainControl(method = "repeatedcv", 
                     number = 6, repeats = 3)

rf_mood <- train(cluster ~., 
                 mood_train, 
                 method = "rf",
                 trControl = ctrl)
# predict
rf_pred <- predict(rf_mood, mood_test, type = "raw")

# confusion matrix
rf_result <- confusionMatrix(as.factor(rf_pred), as.factor(mood_test$cluster))

df_rf_result <- as.data.frame(as.table(rf_result)) %>% 
  mutate(
    model = 'RANDOM FOREST'
  )

model_result <- rbind(model_result, df_rf_result)

# result for Accuracy
acc_rf <- as.data.frame(as.table(
  as.matrix(rf_result, 
            what = "overall"))) %>% 
  mutate(
    Model = 'RANDOM FOREST'
  ) %>% 
  filter(
    Var1 == 'Accuracy'
  ) %>% 
  rename(
    Accuracy = Freq
  ) %>% 
  select(Model, Accuracy)

accuracy <- rbind(accuracy, acc_rf)

SVM

# preparation

mood_train <- mood_train %>% 
  mutate(cluster = as.factor(cluster))

# build model svm

svm_mood <- svm(
  formula = cluster ~.,
  data = mood_train,
  type = 'C-classification',
  kernel = "linear"
)

# predict with data test
svm_pred <- predict(svm_mood, mood_test)

svm_result <- confusionMatrix(as.factor(svm_pred), 
                            as.factor(mood_test$cluster))

df_svm_result <- as.data.frame(as.table(svm_result)) %>% 
  mutate(
    model = 'SVM'
  )

model_result <- rbind(model_result, df_svm_result)

# result for Accuracy
acc_svm <- as.data.frame(as.table(
  as.matrix(svm_result, what = "overall"))) %>% 
  mutate(
    Model = 'SVM'
  ) %>% 
  filter(
    Var1 == 'Accuracy'
  ) %>% 
  rename(
    Accuracy = Freq
  ) %>% 
  select(Model, Accuracy)

accuracy <- rbind(accuracy, acc_svm)

# show all model performance by accuracy
accuracy$Accuracy <- round(accuracy$Accuracy,2)*100
accuracy

##           Model Accuracy
## 1           KNN       92
## 2   NAIVE BAYES       88
## 3 RANDOM FOREST       84
## 4           SVM       92

Dari hasil akurasi diperoleh bahwa selisih antar kedua metode antara KNN/SVM sudah cukup baik dalam memprediksi mood kita sebagai pendengar.