PROJECT TITLE - Commercial Sales Prediction Analysis using Regression

Dataset Information

The data scientists at BigMart have collected 2013 sales data for 1559 products across 10 stores in different cities. Also, certain attributes of each product and store have been defined. The aim is to build a predictive model and find out the sales of each product at a particular store.

Using this model, BigMart will try to understand the properties of products and stores which play a key role in increasing sales.

MODELS USED

In order to maintain successful business, marketing enterprises like bigmart engaged in wholesaling, manufacturing, logistics and retailing should predict their sales before actual sales. Marketing enterprises account over 10% of the country’s GDP. It is essential for them need to analyze and predict their sales for avoiding unnecessary expenditure. A marketing enterprise should have a model that can predict sales accurately so that customers future demands can be met and accordingly update their sales record. To achieve this, dataset of Big Mart Sales is collected and used for analysis of sales of each product at a particular store. Data exploration, preprocessing, exploratory data analysis and model training methodology is followed for building a predictive model. This paper explores all basic regression algorithms models like linear regression, decision tree, random forest, lasso, ridge and extra-trees regressor for predicting and improving sales with a promising accuracy.

LIBRARIES NEEDED

• numpy
• pandas
• matplotlib
• seaborn

DATASET

https://www.kaggle.com/devashish0507/big-mart-sales-prediction

INTRODUCTION It is often assumed that more sales means more profit. One thing marketing enterprises must have thought when increasing sales and profits is the natural desperation for avoiding loses and unneeded expenditure. That desperation without a proper plan can cause downfall of the entire enterprise. The dataset collected have defined attributes that has numerous dependent and independent variables. Then, the data of the dataset is filtered to get accurate predictions and get some good results that help us to improve knowledge on the data we collected. This knowledge on dataset can further be used for predicting future sales of commercial products by applying machine learning algorithms like random forest and basic linear regression.

LITERATURE SURVEY The BigMart dataset is a collection sales data of 2013. The dataset is divided into 11 input attributes and 1 output attribute. Training on the model is done on 1559 products across 10 stores in different cities.

About the Dataset The Dataset is collected from Kaggle. It has tabular data that corresponds to one or more database tables. Every column of the table represents a particular variable and each row corresponds to a given record of the dataset gathered by the reviews of the customers. Then, the data of the dataset is filtered to get accurate predictions and get some good results that help us to improve knowledge on the data we collected. Sample Data and Basic Analysis on the dataset Screenshots: The dataset has some missing values as other attributes have close by values, thus attributes needs to be processed separately. Unique values in the data set are checked by using the Lambda function. Preprocessing is performed on the dataset. Firstly null values are checked and then categorical attributes are checked. Item and outlet identifier are discarded which increases the accuracy of the model. Further missing values are filled by the mean values and if the values of some attributes are less than 100 we combine them to similar categories and then check the final count. Fig. 1.1 Contents of Dataset Fig. 1.2 Display of Dataset Attributes Fig. 1.3 Dataset Null Values Check Fig. 1.4 Dataset Statistical Information New attributes are created using Lambda function. Old values are mapped to new values. 3 broad categories are the observed. Futher we create one new attribute it is noticed that outlet establishment here is reducing the performance of the model. So new value which is lesser is subtracted from the year 2013 with the following years mentioned in outlet establishment year column.

Hypothesis Generation Data Exploration Data Cleaning Feature Engineering Model Building Hypothesis Generation Understanding the problem better by brainstorming possible factors that can impact the outcome Data Exploration Analysis of the data to uncover the patterns, point of interest and characteristics. Data Cleaning Modification of the data based on missing values identification. Feature Engineering Modifying existing variables and creating new ones for analysis. Model Building Making predictive models on the data. Hypothesis about the data was made and data exploration was performed where some subtle differences in data was found, which required remediation. Next, data cleaning was performed. Then, feature engineering was performed where missing values were imputed and solved the irregularities. New features where included and this made data model friendly by one hot encoding. Finally some basic models were imported to find out which one among them gave the best performance.

IMPLEMENTATION

1. Software To run the regression model, following packages are required.  NumPy  Seaborn  Matplotlib  Pandas  Warnings
2. Exploratory Analysis And Observations with flow chart To understand the correspondence between the target variable and output variable the following graphs are plotted and observed. So quick exploration and visualization is performed on the data using sns.displot function as shown. Fig.3.1 Correspondence between target variable and Item_Weight As it can be seen that in Fig.3.1, there is a dominant peak obtained due to inclusion of mean values that was used to fill the missing values in the particular columns of the dataset. Also, it can be observed that most of the items have values as zero because they don’t have data. Fig.3.2 Correspondence between target variable and Item_Visibility As it can be seen that in Fig.3.2, the inclusion of mean values that was used to fill the missing values in the particular columns of the dataset is also the reason for obtaining dominant peak. Fig.3.3 Correspondence between target variable and Item_MRP As it can be seen that in Fig.3.3, there are four dominant peaks obtained in some intervals due to inclusion of mean values that was used to fill the missing values in the particular columns of the dataset. Fig.3.4 Correspondence between target variable and Item_Outlet_Sales As it can be seen that in Fig.3.4, values are high ranging from 0 to 14000. So normalization by using logarithm transformation is done over the graph and a uniform distribution is obtained as can be seen in Fig.3.5. Fig.3.5 Correspondence between target variable and Item_Outlet Sales Uniform Distribution As a result from Fig.3.5, values are some what minimal, so that the model can predict with less error. Further exploration is done over categorical distributions using countplot function. Fig.3.6 Correspondence between target variable and Item_Fat_Content As it can be seen that in Fig.3.6, most of the products are low fat content type and very less products are non edible type. For this matplotlib.pyplot is imported. Further similar approach is performed on the other attributes as shown. Fig.3.7 Correspondence between target variable and Item_Type Fig.3.8 Correspondence between target variable and Outlet_Establishment_Year Fig.3.9 Correspondence between target variable and Outlet_Size Fig.3.10 Correspondence between target variable and Outlet_Location_Type Fig.3.11 Correspondence between target variable and Outlet_Type

Pseudocode from sklearn.model_selection import cross_val_score from sklearn.metrics import mean_squared_error def train(model, X, y):

model.fit(X, y)

pred = model.predict(X)

cv_score = cross_val_score(model, X, y, scoring='neg_mean_squared_error', cv=5) cv_score = np.abs(np.mean(cv_score))

print("Model Report") print("MSE:",meansquared_error(y,pred)) print("CV Score:", cv_score) from sklearn.linear_model import LinearRegression, Ridge, Lasso model = LinearRegression(normalize=True) train(model, X, y) coef = pd.Series(model.coef, X.columns).sortvalues() coef.plot(kind='bar', title="Model Coefficients") model = Ridge(normalize=True) train(model, X, y) coef = pd.Series(model.coef, X.columns).sortvalues() coef.plot(kind='bar', title="Model Coefficients") model = Lasso() train(model, X, y) coef = pd.Series(model.coef, X.columns).sortvalues() coef.plot(kind='bar', title="Model Coefficients") from sklearn.ensemble import RandomForestRegressor model = RandomForestRegressor() train(model, X, y) coef = pd.Series(model.feature_importances, X.columns).sortvalues(ascending=False) coef.plot(kind='bar', title="Feature Importance") from sklearn.ensemble import ExtraTreesRegressor model = ExtraTreesRegressor() train(model, X, y) coef = pd.Series(model.feature_importances, X.columns).sort_values(ascending=False) coef.plot(kind='bar', title="Feature Importance") RESULTS AND DISCUSSION

1. Correlation Matrix Fig.4 Correlation Matrix As it can be seen that in Fig.4, a correlation matrix is created that is displayed in a heat map. A particular color map can be chosen to display it. Following Observations are made:  Item_Weight and Outlet_Establishment_Year are having good correlation.  However, there is a clear difference between Outlet_Establishment_Year and Outlet_Years. They are having negative correlation.  Form Item_Outlet_Sales we get to know which is highly impacting the prediction model.  Therefore, Item_MRP gives the major impact to the outlet sales. Label encoding and One-hot encoding are performed. Initially, for label encoding, label encoder is imported. Firstly, the outlet identifier is converted into numerical values, based on the specific stores the prediction may be affected. For model improvement we converted categorical attribute into label encoder. We convert these categorical attributes into numerical values. For getting, more prediction in the model one hot encoding can be performed it will create new column and thus will increase accuracy of the model. It performs better for categorical attributes. Here, dummies keyword is used to represent one hot encoding. As a result, 26 features are obtained. Train test split process is performed on input variable output variable and a separate variable. For this, some attributes are dropped in input and output attributes columns. Cross validation can also be used to automatically split the dataset accordingly. Model training is performed and the following matrix, cross validation score and mean squared error is used. Here, the function having model, input which is denoted by x and output which is denoted by y is used. The training set is then predicted and cross validation is performed. The cross validation score is often by using negative mean squared error function. With this the model is defined.
2. Graphs from Model Training Some basic models are imported to analyze the performance of the training model. Based on the cross validation score, best model is picked. A. Linear Regression Model Fig.5.1 Linear Regression Model Model Report MSE: 0.2880838937683423 CV Score: 0.28921756014145145 B. Ridge Model Fig.5.2 Ridge Model Model Report MSE: 0.42802985605683475 CV Score: 0.4289289251562643 C. Lasso Model Fig.5.3 Lasso Model Model Report MSE: 0.7628688679102087 CV Score: 0.7630789166281843 D.Decision Tree Regressor Model Fig.5.4 Decision Tree Regressor Model Model Report MSE: 5.5534030638578795e-34 CV Score: 0.5787926637055503 E. Random Forest Regressor Model Fig.5.5 Random Forest Regressor Model Model Report MSE: 0.042468217567918444 CV Score: 0.3098775587950489 F. Extra Trees Regressor Model Fig.5.7 Extra Trees Regressor Model Model Report MSE: 1.0418489584965893e-28 CV Score: 0.33222859882716604

CONCLUSION Based on the reports obtained and analyzed on the training model and dataset, linear regression has the top performance. The study was focused on the 11 input attributes and one output attribute thus, obtaining a promising accuracy with the corresponding cv score. Machine learning techniques have wide range of applications. Many types of research and works are being done on it.