How to choose the optimal threshold for imbalanced classification

threshold_tuning.py

# Array for finding the optimal threshold
thresholds = np.arange(0.0, 1.0, 0.0001)
fscore = np.zeros(shape=(len(thresholds)))
print('Length of sequence: {}'.format(len(thresholds)))

# Fit the model
for index, elem in enumerate(thresholds):
    # Corrected probabilities
    y_pred_prob = (y_pred > elem).astype('int')
    # Calculate the f-score
    fscore[index] = f1_score(y_test, y_pred_prob)

# Find the optimal threshold
index = np.argmax(fscore)
thresholdOpt = round(thresholds[index], ndigits = 4)
fscoreOpt = round(fscore[index], ndigits = 4)
print('Best Threshold: {} with F-Score: {}'.format(thresholdOpt, fscoreOpt))

# Plot the threshold tuning
df_threshold_tuning = pd.DataFrame({'F-score':fscore,
                                    'Threshold':thresholds})
df_threshold_tuning.head()

plotnine.options.figure_size = (8, 4.8)
(
    ggplot(data = df_threshold_tuning)+
    geom_point(aes(x = 'Threshold',
                   y = 'F-score'),
               size = 0.4)+
    # Best threshold
    geom_point(aes(x = thresholdOpt,
                   y = fscoreOpt),
               color = '#981220',
               size = 4)+
    geom_line(aes(x = 'Threshold',
                   y = 'F-score'))+
    # Annotate the text
    geom_text(aes(x = thresholdOpt,
                  y = fscoreOpt),
              label = 'Optimal threshold \n for class: {}'.format(thresholdOpt),
              nudge_x = 0,
              nudge_y = -0.10,
              size = 10,
              fontstyle = 'italic')+
    labs(title = 'Threshold Tuning Curve')+
    xlab('Threshold')+
    ylab('F-score')+
    theme_minimal()
)