Question

Write a neural network in python for multiclass classification of an imbalanced dataset that makes a successful model that both trains and evaluates the model and prints the Accuracy, Precision, Recall, and F1 score. Data will be in the form of a CSV file with 600,000 samples (or rows in the CSV) of 15 classes (for y_train and y_test) and 78 input dimensions/features. The imbalance will have some classes that have a low of only 8 samples taken. Your job is the make a neural network to fix those imbalances and optimize the network for the best Accuracy, Precision, Recall, and F1 score for all 15 classes.

Sample data -

22

6

386359

22

20

1912

2665

640

0

86.90909

137.688

976

0

133.25

268.7713

11846.5

108.7072

9423.39

22717.21

122019

5

385442

18354.38

29591.49

122019

212

386353

20334.37

43298.97

161209

10

0

0

0

0

712

648

56.94186

51.76533

0

976

106.4419

207.2919

42969.92

0

0

0

1

0

0

0

0

0

108.9762

86.90909

133.25

0

0

0

0

0

0

22

1912

20

2665

26883

230

16

32

0

0

0

0

0

0

0

0

Class 4

0

0

1.13E+08

3

0

0

0

0

0

0

0

0

0

0

0

0

0.026634

56319965

33.23402

56319988

56319941

1.13E+08

56319965

33.23402

56319988

56319941

0

0

0

0

0

0

0

0

0

0

0

0.026634

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

0

3

0

0

0

-1

-1

0

0

0

0

0

0

56319965

33.23402

56319988

56319941

Class 7

Answer 1

#Importing the numpy to perform Linear Algebraic operations on the data

import numpy as np

#Import pandas library to perform the data preprocessing

import pandas as pd

#importing the Keras deep learning framework of Python

import keras

#Importing the Sequential model from keras

from keras.models import Sequential

#Importing the types of layers in the Neural Network that we are going to have

from keras.layers import Dense

#Importing the train_test_split function which is useful in dividing the dataset into the training and testing data

from sklearn.model_selection import train_test_split

#Importing the StandardScaler function to perform the standardisation/scaling of the data

from sklearn.preprocessing import StandardScaler

#Importing the metries for the performance evaluation of our deep learning model

from sklearn import metrics

def data_preprocessing(data):

#As you have told that there are 78 features hence the 78th index position will be having the class labels

#This is because the indexing starts from 0 in Pandas in Python

X = data.iloc[:, 0:78]

y = data.iloc[:78]

#I have splitted the dataset into a ratio of 80:20 between the train and test you can try other ratios as well

X_train, X_test, y_train, y_test = train_test_split(X, y, test_size = 0.2, random_state = 23)

#Creating an object of StandardScaler

sc = StandardScaler()

#Scaling the data using the StandardScaler() object

X_train = sc.fit_transform(X_train)

X_test = sc.transform(X_test)

return X_train, X_test, y_train, y_test

def train_and_evaluate(X_train, X_test, y_train, y_test):

neural_classifier = Sequential()

#output_dim = number of nuerons in first hidden layer

#init = initializing the weights of the neural network

#input_dim = number of neuron in the input layer = number of input features = 78 in your case

#Actiavtion = activation function that is used in each layer

#Dense is the type of layer

#Input layer

neural_classifier.add(Dense(output_dim = 100, init = 'uniform', activation = 'relu', input_dim = 78))

#First hidden layer

neural_classifier.add(Dense(output_dim = 150, init = 'uniform', activation = 'relu'))

#Second hidden layer

neural_classifier.add(Dense(output_dim = 200, init = 'uniform', activation = 'relu'))

#Third hidden layer

neural_classifier.add(Dense(output_dim = 250, init = 'uniform', activation = 'relu'))

#Fourth hidden layer

neural_classifier.add(Dense(output_dim = 300, init = 'uniform', activation = 'relu'))

#Fifth hidden layer

neural_classifier.add(Dense(output_dim = 350, init = 'uniform', activation = 'relu'))

#Sixth hidden layer

neural_classifier.add(Dense(output_dim = 400, init = 'uniform', activation = 'relu'))

#Seventh hidden layer

neural_classifier.add(Dense(output_dim = 250, init = 'uniform', activation = 'relu'))

#Eighth hidden layer

neural_classifier.add(Dense(output_dim = 300, init = 'uniform', activation = 'relu'))

#Output layer

#output layer has 15 neurons because there are 15 classes in your dataset

#Since it is a multiclass classification problem hence we are using the softmax activation function

#If it was a binary classification problem then we could have used thee sigmoid activation function

neural_classifier.add(Dense(output_dim = 15, init = 'uniform', activation = 'softmax'))

#Optimizer is ADAM which is a kind of optimization algorithm for minimizing the loss on each and every epoch

neural_classifier.compile(optimizer = 'adam', loss = 'categorical_crossentropy', metrics = ['accuracy'])

#Epochs = number of times we will train our network

neural_classifier.fit(X_train, y_train, batch_size = 10, epochs = 100)

#Predicting the labels for the test data

y_pred = neural_classifier.predict(X_test)

#Calculating the accuracy score

accuracy = metrics.accuracy_score(y_test, y_pred)

#Calculating the precision score

precision = metrics.precision_score(y_test, y_pred)

#Calculating the recall score

recall = metrics.recall_score(y_test, y_pred, average='weighted')

#Calculating the f1-score

f1score = metrics.f1_score(y_test, y_pred, average='weighted')

return accuracy, precision, recall, f1score

def main():

data = pd.read_csv('Filename.csv')

X_train, X_test, y_train, y_test = data_preprocessing(data)

accuracy, precision, recall, f1_score = train_and_evaluate(X_train, X_test, y_train, y_test)

print("Accuracy score of the model is :", accuracy)

print("precision score of the model is :", accuracy)

print("Recall score of the model is :", accuracy)

print("f1-score of the model is :", accuracy)

main()

screenshots: