Libraries And Dataset¶

In [111]:
# Data visualisation libraries
import pandas as pd
import numpy as np
import matplotlib.pyplot as plt
import seaborn as sns

# Random Forest utils
from sklearn.ensemble import RandomForestClassifier
from sklearn.pipeline import make_pipeline
from category_encoders import OrdinalEncoder
from sklearn.compose import ColumnTransformer
from sklearn.impute import SimpleImputer
from sklearn.model_selection import GridSearchCV, cross_val_score, train_test_split, RandomizedSearchCV
from sklearn.metrics import accuracy_score, precision_score, recall_score, ConfusionMatrixDisplay, make_scorer, classification_report, confusion_matrix
from sklearn.feature_extraction.text import TfidfVectorizer
from sklearn.preprocessing import LabelEncoder, OneHotEncoder, StandardScaler,MinMaxScaler
In [112]:
dataset = pd.read_csv('data/market_cluster.csv', encoding='latin1')
In [113]:
dataset.head()
Out[113]:
Order ID Customer Name Category Sub Category City Order Date Region Sales Discount Profit State profit_margin Cluster
0 OD1 Harish Oil & Masala Masalas Vellore 11-08-2017 North 1254 0.12 401.28 Tamil Nadu 0.32 Medium
1 OD2 Sudha Beverages Health Drinks Krishnagiri 11-08-2017 South 749 0.18 149.80 Tamil Nadu 0.20 Medium
2 OD3 Hussain Food Grains Atta & Flour Perambalur 06-12-2017 West 2360 0.21 165.20 Tamil Nadu 0.07 Low
3 OD4 Jackson Fruits & Veggies Fresh Vegetables Dharmapuri 10-11-2016 South 896 0.25 89.60 Tamil Nadu 0.10 Low
4 OD5 Ridhesh Food Grains Organic Staples Ooty 10-11-2016 South 2355 0.26 918.45 Tamil Nadu 0.39 High
In [114]:
dataset.describe()
Out[114]:
Sales Discount Profit profit_margin
count 9994.000000 9994.000000 9994.000000 9994.000000
mean 1496.596158 0.226817 374.937082 0.250228
std 577.559036 0.074636 239.932881 0.118919
min 500.000000 0.100000 25.250000 0.050000
25% 1000.000000 0.160000 180.022500 0.150000
50% 1498.000000 0.230000 320.780000 0.250000
75% 1994.750000 0.290000 525.627500 0.350000
max 2500.000000 0.350000 1120.950000 0.450000

Data Cleaning & Preprocessing¶

In [115]:
dataset.drop(['Order ID'], axis=1, inplace=True)
In [116]:
dataset.isna().sum()
Out[116]:
Customer Name    0
Category         0
Sub Category     0
City             0
Order Date       0
Region           0
Sales            0
Discount         0
Profit           0
State            0
profit_margin    0
Cluster          0
dtype: int64
In [117]:
def remove_outliers(data: pd.DataFrame, column: str) -> pd.Series:
    q3, q1 = np.nanpercentile(data[column], [75, 25])
    iqr = q3 - q1
    upper_bound = q3 + 1.5 * iqr
    lower_bound = q1 - 1.5 * iqr
    data = data[(data[column] > lower_bound) & (data[column] < upper_bound)]

    return data

dataset = remove_outliers(dataset, 'Discount')
dataset = remove_outliers(dataset, 'Sales')
dataset = remove_outliers(dataset, 'Profit')
In [118]:
dataset.head()
Out[118]:
Customer Name Category Sub Category City Order Date Region Sales Discount Profit State profit_margin Cluster
0 Harish Oil & Masala Masalas Vellore 11-08-2017 North 1254 0.12 401.28 Tamil Nadu 0.32 Medium
1 Sudha Beverages Health Drinks Krishnagiri 11-08-2017 South 749 0.18 149.80 Tamil Nadu 0.20 Medium
2 Hussain Food Grains Atta & Flour Perambalur 06-12-2017 West 2360 0.21 165.20 Tamil Nadu 0.07 Low
3 Jackson Fruits & Veggies Fresh Vegetables Dharmapuri 10-11-2016 South 896 0.25 89.60 Tamil Nadu 0.10 Low
4 Ridhesh Food Grains Organic Staples Ooty 10-11-2016 South 2355 0.26 918.45 Tamil Nadu 0.39 High
In [119]:
sns.histplot(dataset['Cluster'])
Out[119]:
<AxesSubplot:xlabel='Cluster', ylabel='Count'>
In [120]:
encoder = LabelEncoder()
scaler = StandardScaler()
onehot = OneHotEncoder()
minmaxscaler = MinMaxScaler()
In [121]:
dataset["Order Date"] = pd.to_datetime(dataset["Order Date"])
dataset["Order Date"] = dataset["Order Date"].dt.month

dataset["Customer Name"] = encoder.fit_transform(dataset["Customer Name"])
dataset["Category"] = encoder.fit_transform(dataset["Category"])
dataset["City"] = encoder.fit_transform(dataset["City"])
dataset["Region"] = encoder.fit_transform(dataset["Region"])
dataset["State"] = encoder.fit_transform(dataset["State"])
dataset["Sub Category"] = encoder.fit_transform(dataset["Sub Category"])

# dataset["Customer Name"] = onehot.fit_transform(dataset["Customer Name"].values.reshape(-1, 1)).toarray()
# dataset["Category"] = onehot.fit_transform(dataset["Category"].values.reshape(-1, 1)).toarray()
# dataset["City"] = onehot.fit_transform(dataset["City"].values.reshape(-1, 1)).toarray()
# dataset["Region"] = onehot.fit_transform(dataset["Region"].values.reshape(-1, 1)).toarray()
# dataset["State"] = onehot.fit_transform(dataset["State"].values.reshape(-1, 1)).toarray()
# dataset["Sub Category"] = onehot.fit_transform(dataset["Sub Category"].values.reshape(-1, 1)).toarray()

# dataset["Order Date"] = encoder.fit_transform(dataset["Order Date"])
In [122]:
# dataset[["Sales", "Discount", "profit_margin","Profit"]] = scaler.fit_transform(dataset[["Sales", "Discount", "profit_margin","Profit"]])
In [123]:
class_to_numeric = {'Low': 0, 'Medium': 1, 'High': 2}
dataset['Cluster'] = [class_to_numeric[label] for label in dataset['Cluster']]
In [124]:
dataset.head()
Out[124]:
Customer Name Category Sub Category City Order Date Region Sales Discount Profit State profit_margin Cluster
0 12 5 14 21 11 2 1254 0.12 401.28 0 0.32 1
1 37 1 13 8 11 3 749 0.18 149.80 0 0.20 1
2 14 3 0 13 6 4 2360 0.21 165.20 0 0.07 0
3 15 4 12 4 10 3 896 0.25 89.60 0 0.10 0
4 28 3 18 12 10 3 2355 0.26 918.45 0 0.39 2

Data Split¶

In [125]:
dataset = dataset.fillna(dataset.mean())

# Check for and handle infinite values
dataset = dataset.replace([np.inf, -np.inf], np.nan)
dataset = dataset.dropna()
In [126]:
X = dataset.drop(['Cluster','Sub Category','State','profit_margin'],axis=1)
y = dataset['Cluster']
In [127]:
X.head()
Out[127]:
Customer Name Category City Order Date Region Sales Discount Profit
0 12 5 21 11 2 1254 0.12 401.28
1 37 1 8 11 3 749 0.18 149.80
2 14 3 13 6 4 2360 0.21 165.20
3 15 4 4 10 3 896 0.25 89.60
4 28 3 12 10 3 2355 0.26 918.45
In [128]:
# from sklearn.feature_extraction.text import TfidfVectorizer
# from sklearn.preprocessing import LabelEncoder
# tfidf_vectorizer = TfidfVectorizer(max_features=1000) 

# X_text = tfidf_vectorizer.fit_transform(dataset['Category'])

# # Example: Encode categorical features using LabelEncoder
# label_encoder = LabelEncoder()
# X_author = label_encoder.fit_transform(dataset['Profit'])
# X = pd.concat([pd.DataFrame(X_text.toarray()), pd.DataFrame(X_author)], axis=1)
In [129]:
# X = dataset.drop(['If_Profit','Sales','profit_margin'],axis=1)
# X = dataset['Sales']
# y = dataset['If_Profit']
In [130]:
X_train_full, X_test, y_train_full, y_test = train_test_split(X, y, test_size=0.2, stratify= y) 
X_train, X_valid, y_train, y_valid = train_test_split(X_train_full, y_train_full, test_size= 0.2)

Random Forest Classifier¶

In [131]:
rfc = RandomForestClassifier(random_state=42)

params = { 
    "n_estimators": range(50,350,50), 
    "max_depth": range(5,41,5), 
    "criterion": ["gini", "entropy"]
}
In [132]:
model_rf =  GridSearchCV(estimator=rfc, #
    param_grid=params,
    cv=2, 
    n_jobs=-1,
    verbose=1
)
In [133]:
model_rf.fit(X_train, y_train) # model fit

Fitting 2 folds for each of 96 candidates, totalling 192 fits
Out[133]:
GridSearchCV(cv=2, estimator=RandomForestClassifier(random_state=42), n_jobs=-1,
             param_grid={'criterion': ['gini', 'entropy'],
                         'max_depth': range(5, 41, 5),
                         'n_estimators': range(50, 350, 50)},
             verbose=1)
In a Jupyter environment, please rerun this cell to show the HTML representation or trust the notebook.
On GitHub, the HTML representation is unable to render, please try loading this page with nbviewer.org.
GridSearchCV(cv=2, estimator=RandomForestClassifier(random_state=42), n_jobs=-1,
             param_grid={'criterion': ['gini', 'entropy'],
                         'max_depth': range(5, 41, 5),
                         'n_estimators': range(50, 350, 50)},
             verbose=1)
RandomForestClassifier(random_state=42)
RandomForestClassifier(random_state=42)
In [140]:
import joblib
joblib.dump(model_rf, 'model/randomforest_model.pkl')
Out[140]:
['model/randomforest_model.pkl']
In [135]:
cv_results_rf = pd.DataFrame(model_rf.cv_results_)
cv_results_rf.sort_values("rank_test_score").head(1)
Out[135]:
mean_fit_time std_fit_time mean_score_time std_score_time param_criterion param_max_depth param_n_estimators params split0_test_score split1_test_score mean_test_score std_test_score rank_test_score
77 3.911072 0.190646 0.19808 0.004832 entropy 25 300 {'criterion': 'entropy', 'max_depth': 25, 'n_e... 0.949121 0.940641 0.944881 0.00424 1
In [136]:
y_pred = model_rf.predict(X_test)
In [137]:
accuracy = accuracy_score(y_test, y_pred)
print(f"Accuracy : {accuracy:.2f}")

Accuracy : 0.96
In [138]:
cm = confusion_matrix(y_test, y_pred)
disp = ConfusionMatrixDisplay(confusion_matrix=cm)
disp.plot(cmap='Blues', values_format='d')
plt.title("Confusion Matrix: Random Forest")
plt.show()
In [139]:
print(classification_report(y_test, y_pred))

              precision    recall  f1-score   support

           0       0.98      0.98      0.98       682
           1       0.94      0.95      0.95       679
           2       0.97      0.96      0.97       630

    accuracy                           0.96      1991
   macro avg       0.97      0.96      0.96      1991
weighted avg       0.96      0.96      0.96      1991