# Upload .csv file in google colab
from google.colab import files

uploaded = files.upload()

Upload widget is only available when the cell has been executed in the current browser session. Please rerun this cell to enable.

Saving employee_churn_data.csv to employee_churn_data (3).csv

# Import all dependencies
from sklearn.model_selection import train_test_split
from sklearn.preprocessing import StandardScaler
import pandas as pd
import tensorflow as tf
import matplotlib.pyplot as plt
import seaborn as sns

#  Import and read the csv into a DataFrame
import pandas as pd
employees_df = pd.read_csv("employee_churn_data.csv")
employees_df

	department	promoted	review	projects	salary	tenure	satisfaction	bonus	avg_hrs_month	left
0	operations	0	0.577569	3	low	5.0	0.626759	0	180.866070	no
1	operations	0	0.751900	3	medium	6.0	0.443679	0	182.708149	no
2	support	0	0.722548	3	medium	6.0	0.446823	0	184.416084	no
3	logistics	0	0.675158	4	high	8.0	0.440139	0	188.707545	no
4	sales	0	0.676203	3	high	5.0	0.577607	1	179.821083	no
...	...	...	...	...	...	...	...	...	...	...
9535	operations	0	0.610988	4	medium	8.0	0.543641	0	188.155738	yes
9536	logistics	0	0.746887	3	medium	8.0	0.549048	0	188.176164	yes
9537	operations	0	0.557980	3	low	7.0	0.705425	0	186.531008	yes
9538	IT	0	0.584446	4	medium	8.0	0.607287	1	187.641370	yes
9539	finance	0	0.626373	3	low	7.0	0.706455	1	185.920934	yes

9540 rows × 10 columns

Preprocessing data

# Check data types and null values
employees_df.info()

<class 'pandas.core.frame.DataFrame'>
RangeIndex: 9540 entries, 0 to 9539
Data columns (total 10 columns):
 #   Column         Non-Null Count  Dtype  
---  ------         --------------  -----  
 0   department     9540 non-null   object 
 1   promoted       9540 non-null   int64  
 2   review         9540 non-null   float64
 3   projects       9540 non-null   int64  
 4   salary         9540 non-null   object 
 5   tenure         9540 non-null   float64
 6   satisfaction   9540 non-null   float64
 7   bonus          9540 non-null   int64  
 8   avg_hrs_month  9540 non-null   float64
 9   left           9540 non-null   object 
dtypes: float64(4), int64(3), object(3)
memory usage: 745.4+ KB

# Determine the number of unique values in each column
employees_df_unique = employees_df.nunique()
employees_df_unique

department         10
promoted            2
review           9540
projects            4
salary              3
tenure             11
satisfaction     9540
bonus               2
avg_hrs_month    9540
left                2
dtype: int64

# Check number of employees turnover
left_counts = employees_df['left'].value_counts()
left_counts.plot(kind='bar', color=['skyblue', 'orange'])

# Plot bar graph
plt.title('Count of Employees who Left the company')
plt.ylabel('Count')
plt.show()

> Model Initialization

# Convert categorical data to numeric with `pd.get_dummies`

# Add.astype(int) to switch boolean variables (True/False) to integers
employees_df_dummies = pd.get_dummies(employees_df).astype(int)
employees_df_dummies

	promoted	review	projects	tenure	satisfaction	bonus	avg_hrs_month	department_IT	department_admin	department_engineering	...	department_marketing	department_operations	department_retail	department_sales	department_support	salary_high	salary_low	salary_medium	left_no	left_yes
0	0	0	3	5	0	0	180	0	0	0	...	0	1	0	0	0	0	1	0	1	0
1	0	0	3	6	0	0	182	0	0	0	...	0	1	0	0	0	0	0	1	1	0
2	0	0	3	6	0	0	184	0	0	0	...	0	0	0	0	1	0	0	1	1	0
3	0	0	4	8	0	0	188	0	0	0	...	0	0	0	0	0	1	0	0	1	0
4	0	0	3	5	0	1	179	0	0	0	...	0	0	0	1	0	1	0	0	1	0
...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...
9535	0	0	4	8	0	0	188	0	0	0	...	0	1	0	0	0	0	0	1	0	1
9536	0	0	3	8	0	0	188	0	0	0	...	0	0	0	0	0	0	0	1	0	1
9537	0	0	3	7	0	0	186	0	0	0	...	0	1	0	0	0	0	1	0	0	1
9538	0	0	4	8	0	1	187	1	0	0	...	0	0	0	0	0	0	0	1	0	1
9539	0	0	3	7	0	1	185	0	0	0	...	0	0	0	0	0	0	1	0	0	1

9540 rows × 22 columns

# Dropping duplicate column(s)
employees_df_dropped = employees_df_dummies.drop(['left_no'], axis =1)
employees_df_dropped

	promoted	review	projects	tenure	satisfaction	bonus	avg_hrs_month	department_IT	department_admin	department_engineering	...	department_logistics	department_marketing	department_operations	department_retail	department_sales	department_support	salary_high	salary_low	salary_medium	left_yes
0	0	0	3	5	0	0	180	0	0	0	...	0	0	1	0	0	0	0	1	0	0
1	0	0	3	6	0	0	182	0	0	0	...	0	0	1	0	0	0	0	0	1	0
2	0	0	3	6	0	0	184	0	0	0	...	0	0	0	0	0	1	0	0	1	0
3	0	0	4	8	0	0	188	0	0	0	...	1	0	0	0	0	0	1	0	0	0
4	0	0	3	5	0	1	179	0	0	0	...	0	0	0	0	1	0	1	0	0	0
...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...
9535	0	0	4	8	0	0	188	0	0	0	...	0	0	1	0	0	0	0	0	1	1
9536	0	0	3	8	0	0	188	0	0	0	...	1	0	0	0	0	0	0	0	1	1
9537	0	0	3	7	0	0	186	0	0	0	...	0	0	1	0	0	0	0	1	0	1
9538	0	0	4	8	0	1	187	1	0	0	...	0	0	0	0	0	0	0	0	1	1
9539	0	0	3	7	0	1	185	0	0	0	...	0	0	0	0	0	0	0	1	0	1

9540 rows × 21 columns

Split Train/Test

# # Split our data into features(X) and target variable(y)
y = employees_df_dropped ["left_yes"].values
X = employees_df_dropped.drop(columns="left_yes").values

# Split our data into a training and testing dataset
X_train, X_test, y_train, y_test = train_test_split(X, y, random_state=78)

X_train.shape

(7155, 20)

X_test.shape

(2385, 20)

# Create a StandardScaler instances
scaler = StandardScaler()

# Fit the StandardScaler
X_scaler = scaler.fit(X_train)

# Scale the data
X_train_scaled = X_scaler.transform(X_train)
X_test_scaled = X_scaler.transform(X_test)

Compile, Train Model with training dataset, Evaluate Model with testing dataset

# Define the neural network model

nn = tf.keras.models.Sequential()

# First hidden layer
nn.add(tf.keras.layers.Dense(units=40, activation="relu", input_dim=len(X_train[0])))

# Second hidden layer
nn.add(tf.keras.layers.Dense(units=20, activation="relu"))

# Output layer
nn.add(tf.keras.layers.Dense(units=1, activation="sigmoid"))

# Check the structure of the model
nn.summary()

Model: "sequential_4"
_________________________________________________________________
 Layer (type)                Output Shape              Param #   
=================================================================
 dense_12 (Dense)            (None, 40)                840       
                                                                 
 dense_13 (Dense)            (None, 20)                820       
                                                                 
 dense_14 (Dense)            (None, 1)                 21        
                                                                 
=================================================================
Total params: 1681 (6.57 KB)
Trainable params: 1681 (6.57 KB)
Non-trainable params: 0 (0.00 Byte)
_________________________________________________________________

# Compile the model
nn.compile(loss="binary_crossentropy", optimizer="adam", metrics=["accuracy"])

# Train the model
fit_model = nn.fit(X_train_scaled, y_train, epochs=50)

Epoch 1/50
224/224 [==============================] - 1s 2ms/step - loss: 0.6188 - accuracy: 0.6904
Epoch 2/50
224/224 [==============================] - 0s 2ms/step - loss: 0.5971 - accuracy: 0.7078
Epoch 3/50
224/224 [==============================] - 0s 2ms/step - loss: 0.5899 - accuracy: 0.7079
Epoch 4/50
224/224 [==============================] - 0s 2ms/step - loss: 0.5834 - accuracy: 0.7079
Epoch 5/50
224/224 [==============================] - 0s 2ms/step - loss: 0.5779 - accuracy: 0.7089
Epoch 6/50
224/224 [==============================] - 0s 2ms/step - loss: 0.5716 - accuracy: 0.7093
Epoch 7/50
224/224 [==============================] - 0s 2ms/step - loss: 0.5661 - accuracy: 0.7096
Epoch 8/50
224/224 [==============================] - 0s 2ms/step - loss: 0.5610 - accuracy: 0.7115
Epoch 9/50
224/224 [==============================] - 0s 2ms/step - loss: 0.5563 - accuracy: 0.7143
Epoch 10/50
224/224 [==============================] - 0s 2ms/step - loss: 0.5526 - accuracy: 0.7174
Epoch 11/50
224/224 [==============================] - 0s 2ms/step - loss: 0.5488 - accuracy: 0.7206
Epoch 12/50
224/224 [==============================] - 0s 2ms/step - loss: 0.5445 - accuracy: 0.7258
Epoch 13/50
224/224 [==============================] - 0s 2ms/step - loss: 0.5431 - accuracy: 0.7266
Epoch 14/50
224/224 [==============================] - 0s 2ms/step - loss: 0.5403 - accuracy: 0.7245
Epoch 15/50
224/224 [==============================] - 0s 2ms/step - loss: 0.5378 - accuracy: 0.7298
Epoch 16/50
224/224 [==============================] - 0s 2ms/step - loss: 0.5372 - accuracy: 0.7318
Epoch 17/50
224/224 [==============================] - 1s 2ms/step - loss: 0.5349 - accuracy: 0.7315
Epoch 18/50
224/224 [==============================] - 1s 3ms/step - loss: 0.5345 - accuracy: 0.7332
Epoch 19/50
224/224 [==============================] - 1s 3ms/step - loss: 0.5336 - accuracy: 0.7329
Epoch 20/50
224/224 [==============================] - 1s 3ms/step - loss: 0.5313 - accuracy: 0.7321
Epoch 21/50
224/224 [==============================] - 1s 3ms/step - loss: 0.5302 - accuracy: 0.7331
Epoch 22/50
224/224 [==============================] - 0s 2ms/step - loss: 0.5299 - accuracy: 0.7349
Epoch 23/50
224/224 [==============================] - 0s 2ms/step - loss: 0.5282 - accuracy: 0.7384
Epoch 24/50
224/224 [==============================] - 0s 2ms/step - loss: 0.5277 - accuracy: 0.7365
Epoch 25/50
224/224 [==============================] - 0s 2ms/step - loss: 0.5273 - accuracy: 0.7379
Epoch 26/50
224/224 [==============================] - 0s 2ms/step - loss: 0.5270 - accuracy: 0.7372
Epoch 27/50
224/224 [==============================] - 0s 2ms/step - loss: 0.5255 - accuracy: 0.7395
Epoch 28/50
224/224 [==============================] - 0s 2ms/step - loss: 0.5250 - accuracy: 0.7365
Epoch 29/50
224/224 [==============================] - 0s 2ms/step - loss: 0.5249 - accuracy: 0.7388
Epoch 30/50
224/224 [==============================] - 0s 2ms/step - loss: 0.5243 - accuracy: 0.7389
Epoch 31/50
224/224 [==============================] - 0s 2ms/step - loss: 0.5249 - accuracy: 0.7389
Epoch 32/50
224/224 [==============================] - 0s 2ms/step - loss: 0.5237 - accuracy: 0.7396
Epoch 33/50
224/224 [==============================] - 0s 2ms/step - loss: 0.5229 - accuracy: 0.7407
Epoch 34/50
224/224 [==============================] - 0s 2ms/step - loss: 0.5228 - accuracy: 0.7409
Epoch 35/50
224/224 [==============================] - 0s 2ms/step - loss: 0.5231 - accuracy: 0.7417
Epoch 36/50
224/224 [==============================] - 0s 2ms/step - loss: 0.5226 - accuracy: 0.7402
Epoch 37/50
224/224 [==============================] - 0s 2ms/step - loss: 0.5212 - accuracy: 0.7409
Epoch 38/50
224/224 [==============================] - 0s 2ms/step - loss: 0.5213 - accuracy: 0.7398
Epoch 39/50
224/224 [==============================] - 0s 2ms/step - loss: 0.5206 - accuracy: 0.7426
Epoch 40/50
224/224 [==============================] - 0s 2ms/step - loss: 0.5203 - accuracy: 0.7430
Epoch 41/50
224/224 [==============================] - 0s 2ms/step - loss: 0.5202 - accuracy: 0.7435
Epoch 42/50
224/224 [==============================] - 0s 2ms/step - loss: 0.5199 - accuracy: 0.7420
Epoch 43/50
224/224 [==============================] - 0s 2ms/step - loss: 0.5197 - accuracy: 0.7428
Epoch 44/50
224/224 [==============================] - 0s 2ms/step - loss: 0.5195 - accuracy: 0.7448
Epoch 45/50
224/224 [==============================] - 0s 2ms/step - loss: 0.5185 - accuracy: 0.7441
Epoch 46/50
224/224 [==============================] - 1s 2ms/step - loss: 0.5183 - accuracy: 0.7417
Epoch 47/50
224/224 [==============================] - 1s 3ms/step - loss: 0.5181 - accuracy: 0.7434
Epoch 48/50
224/224 [==============================] - 1s 2ms/step - loss: 0.5180 - accuracy: 0.7458
Epoch 49/50
224/224 [==============================] - 1s 2ms/step - loss: 0.5179 - accuracy: 0.7452
Epoch 50/50
224/224 [==============================] - 1s 3ms/step - loss: 0.5176 - accuracy: 0.7426

# Evaluate the model using the test data
model_loss, model_accuracy = nn.evaluate(X_test_scaled,y_test,verbose=2)
print(f"Loss: {model_loss}, Accuracy: {model_accuracy}")

75/75 - 0s - loss: 0.5599 - accuracy: 0.7174 - 212ms/epoch - 3ms/step
Loss: 0.5598913431167603, Accuracy: 0.7174004316329956

# Create a new DataFrame
history_df = pd.DataFrame(fit_model.history)

# Plot the accuracy
history_df.plot(y="accuracy", color="orangered")
plt.ylabel('Accuracy')
plt.xlabel('Epochs')
plt.show()

RESULTS:

• Accuracy = 71%
• Our trained model yields an accuracy score of 71% having retained all orginal features and their values. This score suggests that our model is not severely overfitting or underfitting the data, and is making predictions correctly about 71.8% of the time. But there is room for improvement. We will reexamine our datset for relevant features, and try different variations in designing the neural network in order to optimize our model further.

> Model Optimization

Correlation Matrix to re-examine features

# Convert categorical features to numerical (only run once! otherwise clear all outputs and rerun)

# Assign a numerical value to each categorical feature instead
department_encoding = {'sales': 0, 'retail': 1, 'operations': 2, 'engineering': 3, 'marketing': 4, 'support': 5, 'admin': 6, 'finance': 7, 'logistics': 8, 'IT': 9}
salary_encoding = {'low': 0, 'medium': 1, 'high': 2}
left_encoding = {'no': 0, 'yes': 1}

# Encoded
employees_df['department_encoded'] = employees_df['department'].map(department_encoding).astype(int)
employees_df['salary_encoded'] = employees_df['salary'].map(salary_encoding).astype(int)
employees_df['left_encoded'] = employees_df['left'].map(left_encoding).astype(int)

# Drop the original categorical columns
employees_df.drop(['department', 'salary', 'left'], axis=1, inplace=True)

# Keep only numeric columns
numeric_df = employees_df.select_dtypes(include=['float64', 'int64'])

# Display correlation matrix
correlation_matrix = numeric_df.corr()

plt.figure(figsize=(8, 6))
sns.heatmap(correlation_matrix, annot=True, cmap='crest', fmt=".2f", linewidths=0.5)
plt.title('Correlation Matrix')
plt.show()

Based on Correlation Matrix, we observe these relationships:

1. 'left' and 'review' have a moderately positive correlation;
2. 'left' and 'promoted'/'projects'/'satisfaction'/'bonus' have weakly negative correlations;
3. 'tenure' and 'avg_hrs_month' have strongly positive correlation themselves;
4. 'department' and 'salary' have no correlation to 'left'.

Let's explore each one more closely:

#1.'left' and 'review' have a moderately positive correlation:
# means there is some tendency for the number of employees who left to increase as the reviews become more positive, possibly for better opportunities.
g = sns.FacetGrid(numeric_df, col='left_encoded')
g.map(plt.hist, 'review', bins=20)

<seaborn.axisgrid.FacetGrid at 0x7a3070976980>

#2. 'left' and 'promoted'/'projects'/'satisfaction'/'bonus' have negative correlations:
# means changes in any one of these factors are not largely associated with consistent changes in the number of employees who left the company.
columns_to_plot = ['promoted', 'projects', 'satisfaction',	'bonus']

# Create loop to go through 4 feature columns
for column in columns_to_plot:
    g = sns.FacetGrid(numeric_df, col='left_encoded')
    g.map(plt.hist, column, bins=10)

plt.show()

#3. 'tenure' and 'avg_hrs_month' have strongly positive correlation:
# means that employees who have both left or stayed with the company for a longer time, between 4.0-9.0 years, tend to work more hours per month on average, or vice versa.
# Keep these columns in our data.

g = sns.FacetGrid(numeric_df, col='left_encoded', hue='tenure')
g.map(plt.hist, 'avg_hrs_month', bins=10)
g.add_legend()

plt.show()

#4. 'department' and 'salary' have no correlation to 'left':
# means neither the department in which an employee works nor their salary level has a significant impact on whether they left the company. Let's consider dropping these two columns.

g = sns.FacetGrid(numeric_df, col='left_encoded', hue='salary_encoded')
g.map(plt.hist, 'department_encoded', bins=20)
g.add_legend()

plt.show()

# Dropping non-essential columns
employees_df_cleaned = numeric_df.drop(['department_encoded', 'salary_encoded'], axis =1)
employees_df_cleaned

	promoted	review	projects	tenure	satisfaction	bonus	avg_hrs_month	left_encoded
0	0	0.577569	3	5.0	0.626759	0	180.866070	0
1	0	0.751900	3	6.0	0.443679	0	182.708149	0
2	0	0.722548	3	6.0	0.446823	0	184.416084	0
3	0	0.675158	4	8.0	0.440139	0	188.707545	0
4	0	0.676203	3	5.0	0.577607	1	179.821083	0
...	...	...	...	...	...	...	...	...
9535	0	0.610988	4	8.0	0.543641	0	188.155738	1
9536	0	0.746887	3	8.0	0.549048	0	188.176164	1
9537	0	0.557980	3	7.0	0.705425	0	186.531008	1
9538	0	0.584446	4	8.0	0.607287	1	187.641370	1
9539	0	0.626373	3	7.0	0.706455	1	185.920934	1

9540 rows × 8 columns

Split Train/Test

# Split our data into features(X) and target variable(y)
y = employees_df_cleaned ["left_encoded"].values
X = employees_df_cleaned.drop(columns="left_encoded").values

# Split our data into a training and testing dataset
X_train, X_test, y_train, y_test = train_test_split(X, y, random_state=78)

X_train.shape

(7155, 7)

X_test.shape

(2385, 7)

# Create a StandardScaler instances
scaler = StandardScaler()

# Fit the StandardScaler
X_scaler = scaler.fit(X_train)

# Scale the data
X_train_scaled = X_scaler.transform(X_train)
X_test_scaled = X_scaler.transform(X_test)

Compile, Train Model with training dataset, Evaluate Model with testing dataset (cleaned)

# Define the neural network model

nn = tf.keras.models.Sequential()

# First hidden layer
nn.add(tf.keras.layers.Dense(units=14, activation="relu", input_dim=len(X_train[0])))

# Second hidden layer
nn.add(tf.keras.layers.Dense(units=7, activation="relu"))

# Output layer
nn.add(tf.keras.layers.Dense(units=1, activation="sigmoid"))

# Check the structure of the model
nn.summary()

Model: "sequential_5"
_________________________________________________________________
 Layer (type)                Output Shape              Param #   
=================================================================
 dense_15 (Dense)            (None, 14)                112       
                                                                 
 dense_16 (Dense)            (None, 7)                 105       
                                                                 
 dense_17 (Dense)            (None, 1)                 8         
                                                                 
=================================================================
Total params: 225 (900.00 Byte)
Trainable params: 225 (900.00 Byte)
Non-trainable params: 0 (0.00 Byte)
_________________________________________________________________

# Compile the model
nn.compile(loss="binary_crossentropy", optimizer="adam", metrics=["accuracy"])

# Train the model
fit_model = nn.fit(X_train_scaled, y_train, epochs=50)

Epoch 1/50
224/224 [==============================] - 2s 3ms/step - loss: 0.5741 - accuracy: 0.7400
Epoch 2/50
224/224 [==============================] - 0s 2ms/step - loss: 0.4801 - accuracy: 0.7553
Epoch 3/50
224/224 [==============================] - 0s 2ms/step - loss: 0.4483 - accuracy: 0.7530
Epoch 4/50
224/224 [==============================] - 0s 2ms/step - loss: 0.4358 - accuracy: 0.7711
Epoch 5/50
224/224 [==============================] - 0s 2ms/step - loss: 0.4277 - accuracy: 0.7955
Epoch 6/50
224/224 [==============================] - 0s 2ms/step - loss: 0.4169 - accuracy: 0.8078
Epoch 7/50
224/224 [==============================] - 0s 2ms/step - loss: 0.4091 - accuracy: 0.8189
Epoch 8/50
224/224 [==============================] - 1s 2ms/step - loss: 0.3988 - accuracy: 0.8280
Epoch 9/50
224/224 [==============================] - 1s 5ms/step - loss: 0.3877 - accuracy: 0.8377
Epoch 10/50
224/224 [==============================] - 1s 5ms/step - loss: 0.3775 - accuracy: 0.8404
Epoch 11/50
224/224 [==============================] - 1s 5ms/step - loss: 0.3698 - accuracy: 0.8451
Epoch 12/50
224/224 [==============================] - 1s 5ms/step - loss: 0.3626 - accuracy: 0.8460
Epoch 13/50
224/224 [==============================] - 1s 5ms/step - loss: 0.3550 - accuracy: 0.8498
Epoch 14/50
224/224 [==============================] - 1s 3ms/step - loss: 0.3484 - accuracy: 0.8510
Epoch 15/50
224/224 [==============================] - 0s 2ms/step - loss: 0.3435 - accuracy: 0.8532
Epoch 16/50
224/224 [==============================] - 1s 3ms/step - loss: 0.3385 - accuracy: 0.8534
Epoch 17/50
224/224 [==============================] - 1s 3ms/step - loss: 0.3350 - accuracy: 0.8559
Epoch 18/50
224/224 [==============================] - 1s 3ms/step - loss: 0.3319 - accuracy: 0.8569
Epoch 19/50
224/224 [==============================] - 1s 3ms/step - loss: 0.3287 - accuracy: 0.8605
Epoch 20/50
224/224 [==============================] - 1s 3ms/step - loss: 0.3257 - accuracy: 0.8625
Epoch 21/50
224/224 [==============================] - 1s 3ms/step - loss: 0.3236 - accuracy: 0.8621
Epoch 22/50
224/224 [==============================] - 0s 2ms/step - loss: 0.3210 - accuracy: 0.8622
Epoch 23/50
224/224 [==============================] - 0s 2ms/step - loss: 0.3191 - accuracy: 0.8637
Epoch 24/50
224/224 [==============================] - 0s 2ms/step - loss: 0.3180 - accuracy: 0.8648
Epoch 25/50
224/224 [==============================] - 0s 2ms/step - loss: 0.3154 - accuracy: 0.8653
Epoch 26/50
224/224 [==============================] - 0s 2ms/step - loss: 0.3139 - accuracy: 0.8668
Epoch 27/50
224/224 [==============================] - 0s 2ms/step - loss: 0.3120 - accuracy: 0.8661
Epoch 28/50
224/224 [==============================] - 1s 5ms/step - loss: 0.3107 - accuracy: 0.8654
Epoch 29/50
224/224 [==============================] - 1s 3ms/step - loss: 0.3088 - accuracy: 0.8657
Epoch 30/50
224/224 [==============================] - 1s 4ms/step - loss: 0.3077 - accuracy: 0.8672
Epoch 31/50
224/224 [==============================] - 0s 2ms/step - loss: 0.3064 - accuracy: 0.8704
Epoch 32/50
224/224 [==============================] - 0s 2ms/step - loss: 0.3047 - accuracy: 0.8693
Epoch 33/50
224/224 [==============================] - 0s 2ms/step - loss: 0.3030 - accuracy: 0.8710
Epoch 34/50
224/224 [==============================] - 0s 2ms/step - loss: 0.3030 - accuracy: 0.8692
Epoch 35/50
224/224 [==============================] - 0s 2ms/step - loss: 0.3020 - accuracy: 0.8690
Epoch 36/50
224/224 [==============================] - 0s 2ms/step - loss: 0.3005 - accuracy: 0.8703
Epoch 37/50
224/224 [==============================] - 0s 2ms/step - loss: 0.2992 - accuracy: 0.8710
Epoch 38/50
224/224 [==============================] - 0s 2ms/step - loss: 0.2994 - accuracy: 0.8721
Epoch 39/50
224/224 [==============================] - 0s 2ms/step - loss: 0.2979 - accuracy: 0.8720
Epoch 40/50
224/224 [==============================] - 0s 2ms/step - loss: 0.2969 - accuracy: 0.8723
Epoch 41/50
224/224 [==============================] - 0s 2ms/step - loss: 0.2961 - accuracy: 0.8741
Epoch 42/50
224/224 [==============================] - 1s 3ms/step - loss: 0.2959 - accuracy: 0.8727
Epoch 43/50
224/224 [==============================] - 1s 3ms/step - loss: 0.2943 - accuracy: 0.8737
Epoch 44/50
224/224 [==============================] - 1s 3ms/step - loss: 0.2940 - accuracy: 0.8734
Epoch 45/50
224/224 [==============================] - 1s 3ms/step - loss: 0.2936 - accuracy: 0.8734
Epoch 46/50
224/224 [==============================] - 1s 3ms/step - loss: 0.2931 - accuracy: 0.8737
Epoch 47/50
224/224 [==============================] - 1s 3ms/step - loss: 0.2920 - accuracy: 0.8728
Epoch 48/50
224/224 [==============================] - 0s 2ms/step - loss: 0.2923 - accuracy: 0.8760
Epoch 49/50
224/224 [==============================] - 0s 2ms/step - loss: 0.2912 - accuracy: 0.8753
Epoch 50/50
224/224 [==============================] - 0s 2ms/step - loss: 0.2907 - accuracy: 0.8762

# Evaluate the model using the test data
model_loss, model_accuracy = nn.evaluate(X_test_scaled,y_test,verbose=2)
print(f"Loss: {model_loss}, Accuracy: {model_accuracy}")

75/75 - 0s - loss: 0.2956 - accuracy: 0.8679 - 218ms/epoch - 3ms/step
Loss: 0.29557913541793823, Accuracy: 0.8679245114326477

# Create a new DataFrame
history_df = pd.DataFrame(fit_model.history)

# Plot the accuracy
history_df.plot(y="accuracy", color="blue")
plt.ylabel('Accuracy')
plt.xlabel('Epochs')
plt.show()

RESULTS 2:

• Accuarcy = 87%

Before finalizing the optimization method above, the following additional attempts were made to improve the model (unsuccessfully):

1. Classification and binning of features such as-

• "review" with cutoff_value = 0.500000, yieliding accuracy=86%;
• "tenure" with cutoff_value = 500 or 4.0 years, yielding accuracy=86%

2. increasing number of neurons to 50/30/1; adding one more hidden layer; activation functions "tanh", yielding accuracy=86%

3. switching to "LeakyReLu"; dropout(0.2), yielding accuracy=84%

In conclusion, dropping "department" and "salary" features proved most effective at enhancing our model performance in predicting future employee turnover.