Data Mining Project

Rossmann Store
Ayari Nadhmi 4 erp-bi 3
Sales

2016-2017
Content

1 Problem Introduction

2 Exploratory data
analysis

3 Models

3 Conclusion

4/29/17 2
2
Problem introduction
Problem Introduction

Forecast sale using store, promotion and competitor data

1115 Stores
Historical Data of sales from January 2013 to July 2015

4/29/17 4
4
Exploratory Data
Analysis
Dataset Details

Train
Variables : store, day of week, date, sales,customers, open, promo,
state holiday, school holiday
Store
store, storetype, assortment, competition distance, competition open since
month, promo2, promo2since week, promo2since year, promo interval

4/29/17 6
6
Dataset Details
Id - an Id that represents a Store
Customers - the number of customers on
a given day Open - an indicator for
whether the store was open: 0 = closed,
1 = open
StateHoliday -indicatesa state holiday.
Normally all stores, with few exceptions,
are closed on state holidays. Note that
all schools are closed on public holidays
and weekends. a = public holiday, b =
Easter holiday, c = Christmas, 0 = None
SchoolHoliday -indicatesif theStore was
affected by the closure of public schools
StoreType-differentiates between 4
different store models: a, b, c, d
PromoInterval -describesthe
consecutive intervals Promo2is started,
naming the months the promotion is
started anew. E.g. "Feb,May,Aug,Nov"
means each round starts in February,
May, August, November of any given
year 4/29/17
for that store
Assortment -describes an assortment
level: a = basic, b = extra,c = extended
CompetitionDistance - distance in meters
to the nearest competitor store
CompetitionOpenSince[Month/Year] -
gives the approximate year and month of
the time the nearest competitor was
opened
Promo - indicates whether a store is
running a promo on that day
Promo2 - Promo2is a continuing and
consecutive promotion for some stores:
0 = store is not participating, 1 = store is
participating
Promo2Since[Year/Week] -describes the
year and calendar week when the store
startedparticipating inPromo2
7
7
Dataset Details

First description

4/29/17 8
8
Dataset Details

Test if there are some stores closed in the training Data

As we can see there no stores closed in the train data

Proportion of open stores against those closed in the train
data

Proportion of Sales against the fact if the store was closed or

open

4/29/17 9
9
Dataset Details
Proportion of the number of customers against closed and
opened stores
As we can see there some days when some stores were open
without having any customers

Proportion of sales against the fact if there have been a promo

or not

4/29/17 10
1
Dataset Details
testing<-train[which(train$Sales!=0 &
train$Customers != 0),]
ggplot(testing,aes(x = factor(testing$Promo), y =
testing$Sales)) +geom_jitter(alpha = 0.1) +
geom_boxplot(color = "yellow", outlier.colour =
NA, fill = NA)

From the graphic we can see the effect of a promo

on Sales

4/29/17 11
1
Dataset Details

testing<-train[which(train$Sales != 0),]
ggplot(testing,aes(x = factor(testing$DayOfWeek),
y = testing$Sales)) +
geom_jitter(alpha = 0.1) +
geom_boxplot(color = "yellow", outlier.colour =
NA, fill = NA)

The plot next to the left shows as the effect of day

on Sales as we can see for days of week 2,3,4 the
sales are mostly the same

4/29/17 12
1
Dataset Details
testing<-train[which(train$Sales != 0),]
ggplot(testing, aes(x =
factor(testing$SchoolHoliday), y =
testing$Sales)) +
geom_jitter(alpha = 0.1, color = "lightblue")
+
geom_boxplot(color = "red", outlier.colour =
NA, fill = NA)

This plot shows as the fact of having a

SchoolHoliday on Sales Amount

4/29/17 13
1
Dataset Details
testing<-train[which(train$Sales != 0 &
train$Customers != 0),]
ggplot(testing,
aes(x = factor(testing$Promo), y =
testing$Customers)) +
geom_jitter(alpha = 0.1, color = "lightblue") +
geom_boxplot(color = "hotpink", outlier.colour =
NA, fill = NA)

The fact of having a promo on the the number of

customers

4/29/17 14
1
Data Conversion and
preprocessing

15
Data Conversion

train$Date<-as.Date(train$Date)
train$month <- as.integer(format(train$Date, "%m"))
train$year <- as.integer(format(train$Date, "%y"))
train$day <- as.integer(format(train$Date, "%d"))
train$SchoolHoliday<-as.factor(train$SchoolHoliday)
train$Promo<-as.factor(train$Promo)

4/29/17 16
1
Data Preprocessing

train_store <- merge(train, store, by = "Store")

train_store <- train_store[train_store$Open != 0, ]

set.seed(123)
trainsample<- sample(1:nrow(train_store),
0.7*nrow(train_store))
test <- sample(setdiff(seq_len(nrow(train_store)),
trainsample), 0.3*nrow(train_store))

4/29/17 17
1
Models Used

18
 Linear regression Model

lr.model <- lm(Sales ~ Promo +DayOfWeek + StateHoliday +month + year +

day+StoreType+ StoreType+CompetitionDistance , train_store[trainsample,])

Model Formula

Residuals with a max

value of 34941

Value being close to 0

indicating that the model
is bad

4/29/17 19
1
Linear regression Model

The rmse given from the above model used We try to improve our model using the StepAIC
for variables selection

4/29/17 20
2
Random Forest Model

library(randomForest)
rf <- randomForest(Sales ~Promo
+DayOfWeek + StateHoliday
+month + year + day+StoreType+
StoreType+SchoolHoliday,train_store
[trainsample,], ntree=20)

4/29/17 21
2
Random Forest Model

varImpPlot(rf) plot(rf)
This plot shows the We can see through this
importance of variables plot the evolution of errror
used by the model within the number of trees

4/29/17 22
2
 Random Forest Model

Using h2o library

4/29/17 23
2
Random Forest Model

4/29/17 24
2
SVM Model

For using the svm regression model we took the store 1

because it took a long time to process all data

4/29/17 25
2
SVM Model

model.SVM <- svm(Sales~Promo +DayOfWeek +

StateHoliday +month + year + day+StoreType+
StoreType+SchoolHoliday , train_store1[sample,])
summary(model.SVM)

4/29/17 26
2
SVM Model

4/29/17 27
2
SVM Model

4/29/17 28
2
Conclusion

VVV 29