BUISNESS REPORT

PROJECT- MACHINE LEARNING

PGPDSBA
SUBMITTED BY- DIKSHHA PARIHAR
DATE- 11-09-2022

TABLE OF CONTENTS

Project objective
Assumptions
Exploratory data analysis
▪ Summary of the dataset
▪ Bivariate analysis
Converting object data type into categorical
Splitting the data into train and test data
▪ Dimensions on the train and test data
Model building
Model Prediction
Model evaluation
Conclusion
Recommendation

Problem 1:

You are hired by one of the leading news channels CNBE who wants to analyze recent elections. This survey was
conducted on 1525 voters with 9 variables. You have to build a model, to predict which party a voter will vote for on the
basis of the given information, to create an exit poll that will help in predicting overall win and seats covered by a
particular party.
Data Dictionary

1. vote: Party choice: Conservative or Labour

2. age: in years

3. economic.cond.national: Assessment of current national economic conditions, 1 to 5.

4. economic.cond.household: Assessment of current household economic conditions, 1 to 5.

5. Blair: Assessment of the Labour leader, 1 to 5.

6. Hague: Assessment of the Conservative leader, 1 to 5.

7. Europe: an 11-point scale that measures respondents' attitudes toward European integration. High

scores represent ‘Eurosceptic’ sentiment.

8. political.knowledge: Knowledge of parties' positions on European integration, 0 to 3.

9. gender: female or male.

Project Objective:

The Objective of the report is to explore the dataset "Election_Data.xlsx"in Python (JUPYTER

NOTEBOOK) and generate insights about the dataset. This exploration report will consist of the following:

• Importing the dataset in jupyter notebook.

• Understanding the structure of dataset.

• Exploratory Data analysis

• Graphical exploration

• Prediction using various machine learning models

• Insights from the dataset

Assumptions:

Predictive modelling is the general concept of building a model that can make predictions. Typically, such a model
includes a machine learning algorithm that learns certain properties from a training dataset to make those
predictions. Pattern classification is to assign discrete class labels to observations as outcomes of a prediction.
Machine learning model predictions allow businesses to make highly accurate guesses as to the likely outcomes
of a question based on historical data, which can be about all kinds of things. These provide the business with
insights that result in tangible business value.

1.1 Read the dataset. Do the descriptive statistics and do the null value condition check. Write an

inference on it.

Dataset: Election_Data.xls
SR.NO unnamed: 0 vote age economic.cond.national economic.cond.household Blair Hague Europe political.knowledge gender
0 1 Labour 43 3 3 4 1 2 2 female

1 2 Labour 36 4 4 4 4 5 2 male

2 3 Labour 35 4 4 5 2 3 2 male

3 4 Labour 24 4 2 2 1 4 0 female

4 5 Labour 41 2 2 1 1 6 2 male

Information on dataset:

Inference

➢ The column “Unnamed : 0” is removed from the dataset before proceeding further as its insignificant for the analysis.

➢ There are 1525 rows and 9 columns

➢ Numerical Columns: age, economical_cond_national, economical_cond_household , Blair, Hague, Europe and

political_knowledge.

➢ Non-Numerical Columns: vote and gender.

➢ There are no null values.

Summary of the dataset:

 count unique top freq mean std min 25% 50% 75% max
vote 1525 2 Labour 1063 NaN NaN NaN NaN NaN NaN NaN
age 1525 NaN NaN NaN 54.182295 15.711209 24 41 53 67 93
economic.cond.national 1525 NaN NaN NaN 3.245902 0.880969 1 3 3 4 5
economic.cond.household 1525 NaN NaN NaN 3.140328 0.929951 1 3 3 4 5
Blair 1525 NaN NaN NaN 3.334426 1.174824 1 2 4 4 5
Hague 1525 NaN NaN NaN 2.746885 1.230703 1 2 2 4 5
Europe 1525 NaN NaN NaN 6.728525 3.297538 1 4 6 10 11
political.knowledge 1525 NaN NaN NaN 1.542295 1.083315 0 0 2 2 3
gender 1525 2 female 812 NaN NaN NaN NaN NaN NaN NaN

Inference

➢ Vote and age variable have 2 unique values

➢ vote: Conservative and Labour

➢ Labour count is 1036 which is more than conservative count

➢ gender: Male and Female

➢ Female voters are more than male voters

➢ age is continuous variable.

➢ minimum age of voters is 24

➢ maximum age of voters is 93

➢ Average age of voters is 54

➢ maximum Assessment of current national economic conditions is 5

➢ Average Assessment of current household economic conditions is 3

➢ minimum Assessment of the Labour leader is 1

➢ Hague: average Assessment of the Conservative leader is 2.7 whereas Blair: average Assessment of

the Labour leader is 3.3.

Duplicates:

Number of duplicate rows = 8

Before (1525, 9)
After (1517, 9)

There are 8 duplicates in the dataframe which was dropped. now there are 1517 rows in the dataframe

Skewness:
Inference

➢ Skewness is a measure of the asymmetry of the probability distribution of a real-valued random

variable about its mean.

➢ age have positive skewness whereas other variables have negative skewness

➢ Blair has more skewness.

Unique values for categorical variables vote: Party choice: Conservative or Labour

Labour 0.69677
Conservative 0.30323
Name: vote, dtype: float64

Inference

➢ nearly 69% vote for labour party only 30% vote for conservative party

Gender

female 0.53263
male 0.46737
Name: gender, dtype: float64
Inference

➢ 53% voters are female voters, and 46% voters are male voters Labour 0.69677.

1.2 Perform Univariate and Bivariate Analysis. Do exploratory data analysis. Check for Outliers.
Inference

➢ only age variable is normally distributed and other variables has multimodal skewness seen.

➢ only economic.cond.national and economic.cond.household have outliers

Bivariate Analysis:

Pair plot between variables:

Inference

➢ There is no linear relationship between variables

➢ Some of the attributes look like they may have an exponential distribution

➢ Conservative party: Knowledge of parties' positions on European integration is unknown

Correlation Heatmap Plot

Inference

➢ There is very less correlation between the variables

➢ The highest positive correlation is seen between “economic_cond_national” and “economic_cond_household” (35%) with
nearly similar results seen from “Blair” and “economic_cond_national” (35%)

➢ The highest negative correlation is seen between “Blair” and “Europe” (29%) with nearly similar results seen from “Blair” and
“Hague” (24%) so, there is less or no chance of multi_collinearity

stripplot between "Hague" and age:

Above plot is a strip plot with jitter as True that really shows the distribution points on the assessment of t he
Conservative leader “Hague” on voters of various age. more voters are distributed in 2 and 4 group.
Catplot Analysis - Blair(count) on economic.cond.national

Assessment of current national economic conditions with Blair shows no 3 cluster have very less distributio n whereas
no 4 cluster have more distribution

Boxplot

Before Outlier removal

After Outlier removal

All outliers are removed for further analysis

1.3 Encode the data (having string values) for Modelling. Is Scaling necessary here or not? Data Split: Split the data
into train and test (70:30).

Converting Object variables to categorical variables:

Inference

➢ Codes are an array of integers which are the positions of the actual values in the categories array.

➢ Here vote and gender are categorical variables are now converted into integers using codes

➢ All the variables in the data frame are integers

Scaling of the data

➢ Most of the times, your dataset will contain features highly varying in magnitudes, units, and range But since, most of
the machine learning algorithms use Euclidean distance between two data points in their computations, this is a
problem.

➢ Differences in the scales across input variables may increase the difficulty of the problem being

modelled.

➢ This means that you are transforming your data so that it fits within a specific scale, like 0-100 or

0-1

➢ Usually, the distance-based methods (E.g.: KNN) would require scaling as it is sensitive to extreme d
ifference and can cause a bias.

➢ tree-based method uses split method (E.g.: Decision Trees) would not require scaling in general as it

s unnecessary

➢ In this dataset, age is only continuous variable and rest of the variables have 1 to 5. Age variable is

only scaled because it is continuous variable

➢ The method of scaling performed only on the ‘age’ variable is the Z-score scaling.

➢ Z-score scaling is the most common form of scaling that takes from the formula (x – mean) / standard deviation).

All the model prediction are done with scaled data.

Train-Test Split

Separating independent (train) and dependent (test)variables for the linear regression model

X = independent (train) variables

Y = dependent (test)variables
The training set for the independent variables: (1061, 8)
The training set for the dependent variable: (1061,)
The test set for the independent variables: (456, 8)
The test set for the dependent variable: (456,)
Inference

spilting the dataset into train and test set to build Logistic regression and LDA model (70:30)

X_train :70% of data randomly chosen from the 8 columns. These are training independent

variables

X_test :30% of data randomly chosen from the 8 columns. These are test independent

variables

y_train :70% of data randomly chosen from the "vote" column. These are training

dependent variables

y_test :30% of data randomly chosen from the "vote" columns. These are test dependent

variables

1.4 Apply Logistic Regression and LDA (linear discriminant analysis).

Logistic Regression Model

Logistic regression is a fundamental classification technique.It belongs to the group of linear classifiers and is somewhat
similar to polynomial and linear regression.It is the go-to method for binary classification problems (problems with two
class values).

Two libraries of Logistic regression

1. Sklearn
2. Statsmodel
Here for the model sklearn library is used
Applying GridSearchCV for Logistic Regression

The probabilities on the training set The probabilities on the test set

Fit the model to the training set

We now fit our model to the GridSearchCV for Logistic Regression model by training the model with our
independent variable and dependent variables.
Inference
#Using GridsearchCV, we input various parameters like 'max_iter', 'penalty',solver', 'tol' which will helps us
to find best grid for prediction of the better model
#max_iter is an integer (100 by default) that defines the maximum number of iterations by the solver
during model fitting.
#solver is a string ('liblinear' by default) that decides what solver to use for fitting the model. Other options
are 'newton-cg', 'lbfgs', 'sag', and 'saga'.
#penalty is a string ('l2' by default) that decides whether there is regularization and which approach to use.
Other options are 'l1', 'elasticnet', and 'none'.
#bestgrid:{'max_iter': 1000, 'penalty': 'l2', 'solver': 'saga', 'tol': 1e-05}
#Accuracy score of training data:83.5%
#Accuracy score of test data:83.5%

The coefficients for each of the independent attributes

The coefficient for age is -0.006854878831463495

The coefficient for economic.cond.national is 0.7135968821433287
The coefficient for economic.cond.household is 0.18721149727710462
The coefficient for Blair is 0.6703768181600023
The coefficient for Hague is -0.745697729263501
The coefficient for Europe is -0.183618187995406
The coefficient for political.knowledge is -0.25573979170938455
The coefficient for gender is 0.15672831586338198

Feature Importance Graphs

Feature: 0, Score: -0.23281

Feature: 1, Score: 0.62889
Feature: 2, Score: 0.06344
Feature: 3, Score: 0.60088
Feature: 4, Score: -0.82301
Feature: 5, Score: -0.21159
Feature: 6, Score: -0.32181
Feature: 7, Score: 0.19209
Inference

➢ The coefficients for each of the independent attributes

➢ The sign of a regression coefficient tells you whether there is a positive or negative correlation
between each independent variable the dependent variable. A positive coefficient indicates that as
the value of the independent variable increases, the mean of the dependent variable also tends to
increase. A negative coefficient suggests that as the independent variable increases, the dependent
variable tends to decrease.
➢ economic.cond.national have more positive coeffiecient . A positive coefficient indicates that as the
value of the independent variable increases, the mean of the dependent variable also tends to
increase the vote
Linear Discriminant Analysis
➢ Linear Discriminant Analysis (LDA) is a dimensionality reduction technique which is commonly used
for the supervised classification problems.
➢ It is used for modeling differences in groups i.e., separating two or more classes. It is used to
project the features in higher dimension space into a lower dimension space.
➢ library used in LDA is sklearn

The probabilities on the training set The probabilities on the test set

Applying GridSearchCV for LDA ➢ Using GridsearchCV, we input various parameters like 'max_iter',
'penalty',solver', 'tol
' which will helps us to find best grid for prediction of the better model
➢ max_iter is an integer (100 by default) that defines the maximum number of iteration
s by the solver during model fitting.
➢ solver is a string ('liblinear' by default) that decides what solver to use for fitting the
model. Other options are 'newton-cg', 'lbfgs', 'sag', and 'saga'.
➢ here ‘solver':['svd', 'lsqr', 'eigen'] are used with others parameters has default
➢ ‘svd’: Singular value decomposition (default). Does not compute the covariance matri
x, therefore this solver is recommended for data with many features.
➢ ‘lsqr’: Least squares solution. Can be combined with shrinkage or custom covariance
estimator.
➢ ‘eigen’: Eigenvalue decomposition. Can be combined with shrinkage or custom covari
ance estimator.
➢ bestgrid:{'solver': 'svd'}
➢ Training Data Class Prediction with a cut-off value of 0.5
➢ Test Data Class Prediction with a cut-off value of 0.5
➢ Accuracy score of training data:83.4%
➢ Accuracy score of test data:83.3%

1.5 Apply KNN Model and Naïve Bayes Model. Interpret the results.

KNN Model

KNN is a non-parametric and lazy learning algorithm. Non-parametric means there is no assumption for un
derlying data distribution.In KNN, K is the number of nearest neighbors. The number of neighbors is the cor
e deciding factor
KNN has the following basic steps:
➢ Calculate distance
➢ Find closest neighbors
➢ Vote for labels
we will be using popular scikit-learn package.
➢ The k-nearest neighbor algorithm is imported from the scikit-learn package.
➢ Create feature and target variables.
➢ Split data into training and test data.
➢ Generate a k-NN model using neighbors value.
➢ Train or fit the data into the model.
➢ Predict the future.

KNN classifier model.

➢ First, import the KNeighborsClassifier module and create KNN classifier object by passing argument
number of neighbors in KNeighborsClassifier() function.
➢ Then, fit your model on the train set using fit() and perform prediction on the test set using predict().
➢ Let us build KNN classifier model for k=15.

misclassification error

[0.2149122807017544,
0.1864035087719298,
0.17763157894736847,
0.18201754385964908,
0.17763157894736847,
0.17324561403508776,
0.17324561403508776,
0.16885964912280704,
0.17105263157894735,
0.16885964912280704]
Plot misclassification error vs k (with k value on X-axis)
The number of neighbors(K) in KNN is a hyperparameter that you need choose at the time of model buildin g.
You can think of K as a controlling variable for the prediction model. n_neighbors = 15

The probabilities on the training set The probabilities on the test set

Accuracy score of training data:84.7%

Accuracy score of test data:83.1%
Gaussian Naive Bayes
Naive Bayes is a classification technique based on the Bayes theorem. It is a simple but powerful
algorithm for predictive modelling under supervised learning algorithms.
Gaussian Naive Bayes – This is a variant of Naive Bayes which supports continuous values and has
an assumption that each class is normally distributed. All we would have to do is estimate the mean and
standard deviation of the continuous variable.
here we use library scikit-learn
Training Data Class Prediction with a cut-off value of 0.5
Test Data Class Prediction with a cut-off value of 0.5

The probabilities on the training set The probabilities on the test set
Accuracy score of training data:83.5%
Accuracy score of test data:82.2%

1.6 Model Tuning, Bagging (Random Forest should be applied for Bagging) and Boosting.

Model tuning

Tuning is the process of maximizing a model’s performance without overfitting or creating too high of a variance.
In machine learning, this is accomplished by selecting appropriate “hyperparameters.” Hyperparameters can be
thought of as the “dials” or “knobs” of a machine learning model. Choosing an appropriate set of
hyperparameters is crucial for model accuracy, but can be computationally challenging. Hyperparameters differ
from other model parameters in that they are not learned by the model automatically through training
methods. Instead, these parameters must be set manually. Many methods exist for selecting appropriate
hyperparameters

Grid Search

Grid Search, also known as parameter sweeping, is one of the most basic and traditional methods of
hyperparametric optimization. This method involves manually defining a subset of the hyperparametric space
and exhausting all combinations of the specified hyperparameter subsets. Each combination’s performance is
then evaluated, typically using cross-validation, and the best performing hyperparametric combination is
chosen.

Bagging with randomforest

A Bagging classifier.

A Bagging classifier is an ensemble meta-estimator that fits base classifiers each on random subsets
of the original dataset and then aggregate their individual predictions (either by voting or by averaging) to
form a final prediction. Such a meta-estimator can typically be used to reduce the variance of a black-box
estimator (e.g., RandomForest), by introducing randomization into its construction procedure and then
making an ensemble out of it.
Bagging and random forests are “bagging” algorithms that aim to reduce the complexity of models that
overfit the training data.
Bagging (Random Forest should be applied for Bagging)
Inference
set the hyper parameters in randomforest classifier
N_estimators (only used in Random Forests) is the number of decision trees used in making the forest
(default = 100).
Max_depth is an integer that sets the maximum depth of the tree. The default is None, which means the
nodes are expanded until all the leaves are pure
Min_samples_split is the minimum number of samples required to split an internal node.

Min_samples_leaf defines the minimum number of samples needed at each leaf. The default input here is
1.
We now fit randomforest classifier model to the bagging model by training the model with our
independent variable and dependent variables.
At this point, you have the classification model defined
The probabilities on the training set The probabilities on the test set

Accuracy score of training data:83.6%

Accuracy score of test data:82.2%

Boosting
Boosting is an ensemble strategy that is consecutively builds on weak learners in order to generate one final
strong learner. A weak learner is a model that may not be exactly accurate or may not take many predictors into
account. By building a weak model, making conclusions about the various feature importance’s and parameters,
and then using those conclusions to build a new, stronger model, Boosting can effectively convert weak learners
into a strong learner.

AdaBoost (Adaptive Boosting):

AdaBoost uses decision stumps as weak learners. A Decision Stump is a Decision Tree model that only splits
off at one level, ergo the final prediction is based off only one feature. When AdaBoost makes its first
Decision Stump, all observations are weighted evenly.
To correct previous error, when moving to the second Decision Stump, the observations that were
classified incorrectly now carry more weight than the observations that were correctly classified. AdaBoost
continues this strategy until the best classification model is built.
➢ GridSearchCV ADA boosting
➢ Using GridsearchCV, we input various parameters like {'algorithm', 'learning_rate', 'n_estimators'}
which will helps us to find best grid for prediction of the better model
➢ N_estimators is the maximum number of estimators at which boosting is terminated. If a perfect fit
is reached, the algo is stopped. The default here is 50.
➢ Learning_rate is the rate at which we are adjusting the weights of our model with respect to the
loss gradient.
➢ The SAMME.R algorithm typically converges faster than SAMME, achieving a lower test error with
fewer boosting iterations.
➢ bestgrid: {'algorithm': 'SAMME.R', 'learning_rate': 0.3, 'n_estimators': 51}

The probabilities on the training set The probabilities on the test set’

Gradient Boosting

Gradient Boosting for classification.

GB builds an additive model in a forward stage-wise fashion; it allows for the optimization of arbitrary
differentiable loss functions. In each stage n_classes_ regression trees are fit on the negative gradient of the
binomial or multinomial deviance loss function. Binary classification is a special case where only a single
regression tree is induced

➢ Using GridsearchCV, we input various parameters like {'criterion', 'loss', 'max_features',

'min_samples_split', 'n_estimators} which will helps us to find best grid for prediction of the better
model
➢ loss{‘deviance’, ‘exponential’}, default=’deviance’
The loss function to be optimized. ‘deviance’ refers to deviance (= logistic regression) for
classification with probabilistic outputs.
➢ criterion{‘friedman_mse’, ‘mse’, ‘mae’}, default=’friedman_mse’
➢ The function to measure the quality of a split. Supported criteria are ‘friedman_mse’ for the mean
squared error with improvement score by Friedman, ‘mse’ for mean squared error, and ‘mae’ for
the mean absolute error.
➢ n_estimatorsint, default=100
➢ The number of boosting stages to perform. Gradient boosting is fairly robust to over-fitting so a
large number usually results in better performance.
➢ min_samples_splitint or float, default=2
➢ The minimum number of samples required to split an internal node:If int, then
consider min_samples_split as the minimum number.
➢ max_features{‘auto’, ‘sqrt’, ‘log2’}, int or float, default=None
➢ The number of features to consider when looking for the best split:
➢ If int, then consider max_features features at each split.
➢ best_params: {'criterion': 'friedman_mse', 'loss': 'exponential', 'max_features': 6,
'min_samples_split': 30, 'n_estimators': 51}
➢ best_estimator:GradientBoostingClassifier(loss='exponential', max_features=6,
min_samples_split=30, n_estimators=51

The probabilities on the training set The probabilities on the test set

➢ Accuracy score of training data:87.4%

➢ Accuracy score of test data:83.3%

1.7 Performance Metrics: Check the performance of Predictions on Train and Test sets using Accuracy,
Confusion Matrix, Plot ROC curve and get ROC_AUC score for each model. Final Model: Compare the models
and write inference which model is best/optimized.
Logistic Regression Model
Confusion matrix on the training and test data

Inference

Training data:
True Negative : 196 False Positive : 111
False Negative : 68 True Positive : 686
Test data:
True Negative : 113 False Positive : 40
False Negative : 35 True Positive : 268

Classification Report of training and test data

AUC and ROC for the training and test data

 Inference

Train Data:

➢ AUC: 89%

➢ Accuracy: 83%

➢ precision : 86%

➢ recall : 91%

➢ f1 :88%

Test Data:

➢ AUC: 88.3%

➢ Accuracy: 84%

➢ precision: 87%

➢ recall : 88%

➢ f1 : 88%

➢ Training and Test set results are almost similar, this proves no overfitting or underfitting

Linear Discriminant Analysis

Confusion matrix on the training and test datax

Inference:

➢ Trainig data:

➢ True Negative : 200 False Positive : 107

➢ False Negative : 69 True Positive : 685

➢ Test data:

➢ True Negative : 111 False Positive : 42

➢ False Negative : 34 True Positive : 269

Classification Report of training and test data

AUC and ROC for the training and test data

AUC for the Training Data: 0.889

AUC for the Test Data: 0.888

Inference

Train Data:

Train Data:

➢ AUC: 88.9%

➢ Accuracy: 83%

➢ precision : 86%

➢ recall : 91%

➢ f1 :89%

Test Data:

➢ AUC: 88.8%

➢ Acuracy: 83%

➢ precision :86%

➢ recall : 89%

➢ f1 : 88%

➢ Training and Test set results are almost similar,This proves no overfitting or underfitting

KNN Model

Confusion matrix on the training and test data

Inference:

Training data:

➢ True Negative : 209 False Positive : 98

➢ False Negative : 64 True Positive : 690

Test data:

➢ True Negative : 116 False Positive : 47

➢ False Negative : 30 True Positive : 273

Classification Report of training and test data

AUC and ROC for the training and test data

AUC for the Training Data: 0.910

AUC for the Test Data: 0.893

Inference

Train Data:

➢ AUC: 91%

➢ Accuracy: 85%

➢ precision : 88%

➢ recall : 92%

➢ f1 :89%

Test Data:

➢ AUC: 89.3%

➢ Accuracy: 83%

➢ precision :85%

➢ recall : 90%

➢ f1 : 88%

➢ Training and Test set results are almost similar,This proves no overfitting or underfitting

Naive Bayes model

Confusion matrix on the training and test data

Inference

Training data:

➢ True Negative : 211 False Positive : 96

➢ False Negative : 79 True Positive : 675

Test data:

➢ True Negative : 112 False Positive : 41

➢ False Negative : 40 True Positive : 263

Classification Report of training and test data

AUC and ROC for the training and test data

AUC for the Training Data: 0.888

Inference

Train Data:

➢ AUC: 88.8%

➢ Accuracy: 84%

➢ precision : 88%

➢ recall : 90%

➢ f1 :89%

Test Data:

➢ AUC: 87.6%

➢ Accuracy: 82%

➢ precision :88%

➢ recall : 88%

➢ f1 : 88%

➢ Training and Test set results are almost similar,This proves no overfitting or underfitting

Bagging with random forest

Confusion matrix on the training and test data

Reference

Training data:

➢ True Negative : 187 False Positive : 120

➢ False Negative : 54 True Positive : 700

Test data:

➢ True Negative : 92 False Positive : 61

➢ False Negative : 20 True Positive : 283

Classification Report of training and test data

AUC and ROC for the training and test data

AUC for the Training Data: 0.897

AUC for the Test Data: 0.884

Inference

Train Data:

➢ AUC: 89.7%

➢ Accuracy: 84%

➢ precision : 85%

➢ recall : 93%

➢ f1 :89%

Test Data:

➢ AUC: 88.4%

➢ Accuracy: 82%

➢ precision :82%

➢ recall : 91%

➢ f1 : 87%

➢ Training and Test set results are almost similar,This proves no overfitting or

underfitting

AdaBoostClassifier

Confusion matrix on the training and test data

Inference

Training data:

➢ True Negative : 197 False Positive : 110

➢ False Negative : 58 True Positive : 696

Test data:

➢ True Negative : 105 False Positive : 48

➢ False Negative : 34 True Positive : 269

Classification Report of training and test data

precision recall f1-score support

0 0.77 0.64 0.70 307

1 0.86 0.92 0.89 754

accuracy 0.84 1061

macro avg 0.82 0.78 0.80 1061

weighted avg 0.84 0.84 0.84 1061

precision recall f1-score support

0 0.76 0.69 0.72 153

1 0.85 0.89 0.87 303

accuracy 0.82 456

macro avg 0.80 0.79 0.79 456

weighted avg 0.82 0.82 0.82 456

AUC and ROC for the training and test data

AUC for the Training Data: 0.906

AUC for the Test Data: 0.889

Inference

Train Data:

➢ AUC: 90.6%

➢ Accuracy: 84%

➢ precision : 85%

➢ recall : 92%

➢ f1 :89%

Test Data:

➢ AUC: 88.9%

➢ Accuracy: 82%

➢ precision :85%

➢ recall : 89%

➢ f1 : 87%

➢ Training and Test set results are almost similar,This proves no overfitting or underfitting

Gradient Boosting

Confusion matrix on the training and test data

Inference

Training data:

➢ True Negative : 224 False Positive : 83

➢ False Negative : 51 True Positive : 703

Test data:

➢ True Negative : 107 False Positive : 46

➢ False Negative : 30 True Positive : 273

Classification Report of training and test data

precision recall f1-score support

0 0.81 0.73 0.77 307

1 0.89 0.93 0.91 754

accuracy 0.87 1061

macro avg 0.85 0.83 0.84 1061

weighted avg 0.87 0.87 0.87 1061

precision recall f1-score support

0 0.78 0.70 0.74 153

1 0.86 0.90 0.88 303

accuracy 0.83 456

macro avg 0.82 0.80 0.81 456

weighted avg 0.83 0.83 0.83 456

AUC and ROC for the training and test data

AUC for the Training Data: 0.934

AUC for the Test Data: 0.901

Inference

Train Data:

➢ AUC: 93.4%

➢ Accuracy: 87%

➢ precision : 89%

➢ recall : 93%

➢ f1 :91%

Test Data:

➢ AUC: 90.1%

➢ Accuracy: 83%

➢ precision :86%

➢ recall : 90%

➢ f1 : 88%

➢ Training and Test set results are almost similar,This proves no overfitting or

underfitting

➢ This Gradient boosting shown better result for this dataset

Final Model: Comparing all the models

Inference

➢ Almost all the models performed well with accuracy between 82% to 84%.

Gradient boosting improved the accuracy to 87% so it is better model for to

predict which party a voter will vote

➢ Comparing all the model ,Gradient boosting model is best model for this

dataset with accuracy of 87% in both training and test set

➢ AUC of Train and test in Gradient boosting model is 93% and 90% respectively

➢ f1 score of Train and test in Gradient boosting model is 91% and 88%

respectively

➢ Precision of Train and test in Gradient boosting model is 89% and 86%

respectively

➢ Recall of Train and test in Gradient boosting model is 93% and 90%

respectively

➢ Accuracy ,AUC,Precision,Recall for test data are almost in line with training

data in Gradient boosting model.This indicates no overfitting or underfitting in

the model

➢ Gradient boosting improved the accuracy to 87% so it is better model for to

predict which party a voter will vote

Overall Best/Optimized model:

Almost all the models performed well with accuracy between 82% to 84% with scaled

data. But Gradient boosting is best and optimised model with accuracy of 87% and also best

AUC,Precision,f1 score, Recall .

LR

Train

LR

Test
LDA

Train

LDA

Test

KNN

Train

KNN

Test

NB

Train

NB

Test

BAGGING

Train

BAGGING

Test

ADA

Train

ADA

Test

Gradient

Train

Gradient

Test

Accuracy 0.83 0.84 0.83 0.83 0.85 0.83 0.84 0.82 0.84 0.82 0.84 0.82 0.87 0.87

AUC 0.89 0.88 0.9 0.88 0.91 0.89 0.89 0.88 0.9 0.88 0.91 0.89 0.93 0.9

Recall 0.91 0.88 0.91 0.89 0.92 0.9 0.9 0.87 0.93 0.93 0.92 0.89 0.93 0.9

Precision 0.86 0.87 0.86 0.86 0.88 0.85 0.88 0.87 0.85 0.82 0.86 0.85 0.89 0.86

F1 Score 0.88 0.88 0.89 0.88 0.89 0.88 0.89 0.87 0.89 0.87 0.89 0.87 0.91 0.88

1.8 Based on these predictions, what are the insights?

The main business objective of this project is to build a model to predict which party a voter will vote for

based on the given information, to create an exit poll that will help in predicting overall win and seats
covered by a particular party.

Various model was built on scaled dataset in that it is found that Gradient Boosting model gave

best/optimized accuracy with 87% to predict which party a voter will vote based on given information and

clearly an exit poll can be built that can help in overall win and seats covered by a particular party.

Sample data test

With sample data we can test against each optimised model.

Information of the sample data:

➢ age' :4

➢ 'economic.cond.national': 6

➢ 'economic.cond.household': 10

➢ 'Blair': 8

➢ 'Hague':20

➢ 'Europe': 11

➢ 'political_knowledge': 7

➢ 'gender': 21

From the sample voters , final model conclusion

Conclusion

Based on this machine learning models, we can predict which party the voter might vote with more sample voters. Exit
polls can be created with this model to predict which party will win or lose and seats covered by a particular party
Problem2 Text analysis on speeches of the Presidents of the United States of America: President Franklin D. Roosevelt
in 1941 President John F. Kennedy in 1961 President Richard Nixon in 1973

In this particular project, we are going to work on the inaugural corpora from the nltk in

Python. We will be looking at the following speeches of the Presidents of the United

States of America:

President Franklin D. Roosevelt in 1941

President John F. Kennedy in 1961

President Richard Nixon in 1973

(Hint: use .words(), .raw(), .sent() for extracting counts)

2.1 Find the number of characters, words, and sentences for the mentioned documents.

characters in Roosevelt speech

The number of characters in Roosevelt speech are: 7571

characters in Kennedy speech

The number of characters in Kennedy speech are: 7618

characters in Nixon speech

The number of characters in Nixon speech are: 9991

Words in Roosevelt speech

The number of Words in Roosevelt speech are: 1536

Words in Kennedy speech

The number of Words in Kennedy speech are: 1546

Words in Nixon speech

The number of Words in Nixon speech are: 2028

sentences in Roosevelt speech

The number of sentences in Roosevelt speech are: 68

sentences in Kennedy speech

The number of sentences in Kennedy speech are: 52

sentences in Nixon speech

The number of sentences in Nixon speech are: 69

2.2 Remove all the stopwords from all three speeches

In natural language processing, useless words (data), are referred to as stop words.

Stop Words:
A stop word is a commonly used word (such as “the”, “a”, “an”, “in”) that a search engine

has been programmed to ignore, both when indexing entries for searching and when

retrieving them as the result of a search query. We would not want these words to take

up space in our database, or taking up valuable processing time. For this, we can remove

them easily, by storing a list of words that you consider to stop words.

Libraries used:

from nltk.corpus import stopwords

nltk.download('stopwords')

Stopwords removal is done through “from nltk.corpus import stopwords”.

Stemming is the process of reducing inflection in words to their root forms such as mapping

a group of words to the same stem even if the stem itself is not a valid word in the

Language.

Library used:

from nltk.stem.porter import PorterStemmer

Most common words in Roosevelt speech after removing stopwords

Most common words in Roosevelt speech after removing stopwords

['nation', 'know', 'peopl', 'spirit', 'life', 'democraci', 'us', 'america', 'live', 'ye

ar', 'human', 'freedom', 'measur', 'men', 'govern', 'new', 'bodi', 'mind', 'speak', 'da

y', 'state', 'american', 'must', 'someth', 'faith', 'unit', 'task', 'preserv', 'within'

, 'histori', 'three', 'form', 'futur', 'seem', 'hope', 'understand', 'thing', 'free', '

alon', 'still', 'everi', 'contin', 'like', 'person', 'world', 'sacr', 'word', 'came', '

land', 'first']

Most common words in Kennedy speech after removing stopwords

Most common words in Kennedy speech after removing stopwords

['let', 'us', 'power', 'world', 'nation', 'side', 'new', 'pledg', 'ask', 'citizen',

'peac', 'shall', 'free', 'final', 'presid', 'fellow', 'freedom', 'begin', 'man', 'h

and', 'human', 'first', 'gener', 'american', 'war', 'alway', 'know', 'support', 'un

it', 'cannot', 'hope', 'help', 'weak', 'arm', 'countri', 'call', 'today', 'well', '

god', 'form', 'poverti', 'life', 'globe', 'right', 'state', 'dare', 'word', 'go', '

friend', 'bear']

Most common words in Nixon speech after removing stopwords

Most common words in Nixon speech after removing stopwords

['us', 'let', 'america', 'peac', 'world', 'respons', 'new', 'nation', 'govern', 'gr

eat', 'year', 'home', 'abroad', 'make', 'togeth', 'shall', 'time', 'polici', 'role'

, 'right', 'everi', 'histori', 'better', 'come', 'respect', 'peopl', 'live', 'help'

, 'four', 'war', 'today', 'era', 'progress', 'other', 'build', 'act', 'challeng', '

one', 'mr', 'share', 'meet', 'promis', 'long', 'work', 'preserv', 'freedom', 'place

', 'system', 'god', 'way']

2.3 Which word occurs the most number of times in his inaugural address for each

president? Mention the top three words. (after removing the stopwords)

Top three words in Roosevelt's speech(after removing the stopwords)

[('nation', 17), ('know', 10), ('peopl', 9)]

Top three words in Kennedy's speech(after removing the stopwords

[('let', 16), ('us', 12), ('power', 9)]

Top three words in Nixon's speech(after removing the stopwords

[('us', 26), ('let', 22), ('america', 21)]

2.4 Plot the word cloud of each of the speeches of the variable. (after removing the

stopwords)

Wordcloud

Many times you might have seen a cloud filled with lots of words in different sizes, which represent the

frequency or the importance of each word. This is called Tag Cloud or WordCloud.

Liabrary used

from wordcloud import WordCloud

Roosevelt speech

Word Cloud (after cleaning)!!

Inference:

➢ Most Frequent words are nation, people, spirit, life

➢ Less Frequent words are weld, kind, women, moment

Kennedy speech
Inference:

➢ Most Frequent words are let, world, slides, power

➢ Less Frequent words are best, wishes, slow

Nixon speech
Inference:

➢ Most Frequent words are America, let, us, nation ➢ Less Frequent words are flimsy, adopted, saw

Conclusion: This project data presented from '1941-Roosevelt.txt', '1961-Kennedy.txt' and '1973-Nixon.txt', we analysed
some interesting insights like the number of characters, words, and sentences from the speeches. To Identify the
strength and the sentiment of these presidential speeches the stop words were removed (punctuation and lowering the
characters were removed) along with stemming. We analysed some of the common words from their speeches which
inspired m