Added R script for Churn Dataset and a simple Python Script for 50_st…

…artups

Lashuk1729_Different_ML_Algorithms/Regression_Startup.py

@@ -0,0 +1,40 @@
+import numpy as np # linear algebra
+import pandas as pd # data processing, CSV file I/O (e.g. pd.read_csv)
+
+import pandas as pd
+import numpy as np
+import matplotlib.pyplot as plt
+from sklearn.linear_model import LinearRegression
+
+from sklearn.compose import ColumnTransformer
+from sklearn.preprocessing import OneHotEncoder
+from sklearn.metrics import accuracy_score, r2_score
+from sklearn.model_selection import cross_val_score
+from sklearn.model_selection import train_test_split
+import seaborn as sns
+
+sns.set_style(style='darkgrid')
+
+# reading the dataset
+dataset = pd.read_csv('../50_Startups.csv')
+x = dataset.iloc[:,:-1].values
+y = dataset.iloc[:,-1].values
+
+# Using One-Hot Encoder
+ct = ColumnTransformer(transformers=[('encoder',OneHotEncoder(),[3])],remainder='passthrough')
+x = np.array(ct.fit_transform(x))
+
+# splitting the data into Training & Test set
+x_train,x_test,y_train,y_test= train_test_split(x,y,test_size=0.2,random_state=0)
+
+# Linear Regression
+lr=LinearRegression()
+lr.fit(x_train,y_train)
+
+# prediction
+y_pred = lr.predict(x_test)
+np.set_printoptions(precision=2)
+print(np.concatenate((y_pred.reshape(len(y_pred),1),y_test.reshape(len(y_test),1)),1))
+
+print(lr.coef_)
+print(lr.intercept_)

Lashuk1729_Different_ML_Algorithms/basic_data_analysis.r

@@ -0,0 +1,267 @@
+#libraries installation
+install.packages('pastecs')
+install.packages('readr')
+install.packages('Jmisc')
+install.packages('ggplot2')
+install.packages('plotly')
+install.packages('corrplot')
+install.packages('caret')
+install.packages('e1071')
+install.packages('gbm')
+install.packages("klaR", dependencies = TRUE)
+install.packages("car", dependencies = TRUE)
+install.packages("mctest")
+install.packages("GGally")
+install.packages("caTools")
+install.packages('mlbench')
+rpart.plot(deciTreeMod, extra = 106)
+
+
+# libraries used
+# for basic data summary
+library(pastecs)
+library(readr)
+
+# for visualization
+library(ggplot2)
+library(plotly)
+library(corrplot)
+library(caTools)
+
+# for checking multicollinearity
+library(mctest)
+library(car)
+library(GGally)
+
+# for applying machine learning algorithms
+# for regression & classification
+library(caret)
+# for svm
+library(e1071)
+# gradient boast
+library(gbm)
+# for classification and visualization(knn, k-modes)
+library(klaR)
+# for decision tree and plotting decision tree
+library(rpart)
+library(rpart.plot)
+# for benchmarking
+library(mlbench)
+
+# reading and previewing first 5 rows of the dataset
+df <- read.csv("../Churn_Modelling.csv")
+head(df, 5)
+
+# checking the information of dataset and number of missing value on each column 
+str(df)
+sapply(df, function(x) sum(is.na(x)))
+
+# checking the unique values for each column
+sapply(df, function(x) length(unique(x)))
+
+# removing first 3 coulumns(features) as they are specific to the customers
+df <- subset(df, select = -c(RowNumber, CustomerId,  Surname))
+dim(df)
+head(df, 2)
+
+# Understanding the y feature of the dataset
+# bar-graph
+ggplot(df, aes(x = Exited)) + 
+  geom_histogram(binwidth = 1, fill = c("blue","Dark Red")) +
+  xlab("Exited") +
+  ylab("Frequency") +
+  ggtitle("Individual who Exited")
+
+# pie-chart
+slices <- table(df$Exited)
+pct <- round(slices/sum(slices)*100)
+lbls<-paste(names(slices),pct,sep=" or ")
+lbls<-paste(lbls,"%")
+pie(slices, labels=lbls, col=terrain.colors(length(lbls)),
+main="Pie Chart of Exited")
+
+# Insight:-
+# only 20% of the customers have currently exited and the goal to model 
+# this feature, more exploration on the data to see the corelation between 
+# some predictors and the outcome variables
+
+# Visualizing how the exited individuals based on tenure
+ggplot(df, aes(x = Exited)) +
+  geom_histogram(aes(x = Tenure,
+                     fill = Exited), binwidth = 1) +
+  geom_vline(aes(xintercept=mean(Tenure))) +
+  facet_grid( ~ Exited)
+
+# Insight:-
+# From the histogram, we can see that average tenure value is around 5 years, 
+# for both exited and not exited individual.
+
+# Visualizing how the exited individuals based on gender
+ggplot(df,aes(x=Gender, y=Exited)) +
+  geom_point(position=position_jitter(0.3))
+
+# Insight:-
+# Considering the plot, we can easily see that Female individuals have exited 
+# more than Male individuals
+
+# Visualizing how the exited individuals based on Geography
+ggplot(df,aes(x=Exited, fill=Geography)) +
+  geom_density(col=NA,alpha=0.25)
+
+# Insight:-
+# From the visualization, we can say that the individuals living in Germany 
+# exited more than those in living in France and Spain.
+
+# Visualizing the correlation between the numerical variables
+p1 <- df[,-which(names(df) == "Geography")]
+p2 <- p1[,-which(names(p1) == "Gender")]
+corr <- cor(p2)
+d <- corrplot(corr)
+
+# Converting Categorical Variables to Numerical and Plotting the new correlation
+# Converting Categorical Variable (Gender) to Numerical
+df$Gender <- as.factor(df$Gender)
+df$Gender <- as.numeric(df$Gender) - 1
+is.numeric(df$Gender)
+
+# Converting Categorical Variable (Geography) to Numerical
+df$Geography <- as.factor(df$Geography)
+df$Geography <- as.numeric(df$Geography) - 1
+is.numeric(df$Geography)
+
+corr <- cor(df)
+d <- corrplot(corr, method="color", order="hclust")
+
+# Insight:-
+# From the correlation plot, we can extract the following insights:-
+# - Age and Balance has some positive correlation with Exited.
+# - Gender and IsActiveMember has some negative correlation with Exited.
+
+
+# After removing and feature engineering, we have the final summary of the 
+# provided dataset
+summary(df)
+str(df)
+
+# splitting the data into Training & Test set
+dataset <- createDataPartition(df$Exited, p = 4/5, list = FALSE)
+df_train <- df[dataset,]
+df_test <- df[-dataset,]
+
+dim(df_train)
+dim(df_test)
+
+# Feature Scaling
+# df_train[c(1,4,5,6,7,10)] <- lapply(df_train[c(1,4,5,6,10)], function(x) c(scale(x)))
+# df_test[c(1,4,5,6,7,10)] <- lapply(df_test[c(1,4,5,6,10)], function(x) c(scale(x)))
+
+# Build the logistic regression model
+logitMod <- glm(Exited ~ ., data=df_train, family=binomial(link="logit"))
+summary(logitMod)
+# Checking Multicollinearity
+vif(logitMod)
+plot(logitMod)
+
+# Insight:
+# Considering the rule of thumb, if VIF is: 
+# 1) 1 = not correlated.
+# 2) Between 1 and 5 = moderately correlated.
+# We can safely say that the features are not correlated.
+
+# Data Preparation
+# Feature Selection: considering relevant and important attributes in your data
+# Ranking the features based on importance
+set.seed(100)
+control <- trainControl(method="repeatedcv", number=10, repeats=3)
+# train the model
+model <- train(Exited~., data=df_train, method="rpart", trControl=control)
+# estimate variable importance
+importance <- varImp(model, scale=FALSE)
+# summarize importance
+print(importance)
+# plot importance
+plot(importance)
+
+# Insight:-
+# Based on the graph, we have see that:-
+# - Features like NumOfProducts, Age, IsActive Member, Balance, Gender and
+#   Geography are first 6 important feature
+# - Other features like EstimatedSalary, CreditScore can be neglected.
+
+################################################################################
+
+# Using the above Logistic Regression Model
+logitMod <- glm(Exited ~ NumOfProducts + Age + IsActiveMember + Balance + 
+                  Gender + Geography, data=df_train, family=binomial)
+
+logPredict <- predict(logitMod, df_test, type="response")
+
+# ROC Curve
+model_AUC <- colAUC(logPredict, df_test$Exited, plotROC = T)
+abline(h = model_AUC, col = "Red")
+text(.2, .9, cex = .8, labels = paste("Optimal Cutoff:", round(model_AUC,4)))
+
+# Cutoff from the ROC Curve = 0.7319
+result_class <- ifelse(logPredict > 0.7319, 1, 0)
+
+# Changing the Exited to binary feature
+result_class <- factor(result_class)
+actual_class <- factor(df_test$Exited)
+
+# Confusion Matrix
+confusionMatrix(result_class, actual_class, mode = "prec_recall", positive="1")
+
+# Insight:-
+# The accuracy for the logistic regression model is 79.5%
+# From the confusion Matrix, we can see that:-
+#   1. Precision : 0.70588         
+#   2. Recall : 0.02878         
+#   3. F1 : 0.05530
+
+################################################################################
+
+# Using Decision Tree
+deciTreeMod <- rpart(Exited ~ NumOfProducts + Age + IsActiveMember + Balance + 
+                       Gender + Geography, data=df_train, method = 'class')
+
+rpart.plot(deciTreeMod, extra = 106)
+
+decitreePredict <- predict(deciTreeMod, df_test, type = "class")
+
+levels(decitreePredict)
+actual_class <- factor(df_test$Exited)
+
+# Confusion Matrix
+confusionMatrix(decitreePredict, actual_class, mode = "prec_recall", positive="1")
+
+# Insight:-
+# The accuracy for the logistic regression model is 84.1%
+# From the confusion Matrix, we can see that:-
+#   1. Precision : 0.7143
+#   2. Recall : 0.3957          
+#   3. F1 : 0.5093
+
+################################################################################
+
+# Model Comparison
+# The main focus is to compare models based on the accuracy. The best model is 
+# Decision Tree with 84.1%, followed by Logistic Regression with 79.1%.
+# I should also consider other algorithms as well for example: Random Forest,
+# Naive Bayes and SVC.
+
+################################################################################
+
+# Conclusion
+# - From the descriptive analysis performed, Female individual exited more 
+#   than Male and Customers located in Germany churn more comparing to 
+#   other locations.
+# - With the Prediction made by the models, Decision Tree seems to classify 
+#   compared to other model with the accuracy of 84.1%.
+# - This accuracy can further be improved by using different algorithms and if 
+#   we are able to obtain additional information for the analysis so that we 
+#   can improve the feature engineering process.
+# - My personal thought is we should concentrate on the individuals who are
+#   exited rather than retaining the new individuals. It's always easier to 
+#   retain the individual who are with us than obtaining new ones. 
+
+################################################################################