Data-Science-Hackathon-Ecommerce-product-categorization-project

Data-Science-Hackathon-by-upGrad

Ecommerce Product Categorization

Project Overview

This project is part of a Data Science Hackathon focused on categorizing ecommerce products based on their descriptions. Using Logistic Regression and TF-IDF vectorization, the model predicts the product category from textual descriptions. The main goal of this project is to classify products into predefined categories using machine learning techniques.

Objectives:

• Clean and preprocess the dataset.
• Convert textual data into numeric form using TF-IDF Vectorization.
• Build and train a Logistic Regression model for multiclass classification.
• Tune hyperparameters using GridSearchCV to optimize the model.
• Evaluate model performance using metrics such as accuracy, precision, recall, F1-score, and confusion matrix.
• Make predictions on unseen data and save the results.

---

Dataset

The dataset used for this project is a collection of ecommerce product descriptions with their associated product categories.

Features:

• description: Textual descriptions of products.
• product_category_tree: Target variable indicating the category of the product.

Preprocessing Steps:

1. Handle Missing Data: Drop the brand column due to a high number of missing values and remove rows with any missing data.
2. Label Encoding: Convert the target variable (product_category_tree) to numerical labels using LabelEncoder.
3. TF-IDF Vectorization: Convert the product descriptions into TF-IDF vectors for numerical representation of text data.

---

Dependencies

The following Python libraries are required:

• pandas
• scikit-learn
• seaborn
• matplotlib

You can install these packages using:

pip install pandas scikit-learn seaborn matplotlib

---

Model Pipeline

1. Data Preprocessing

• Dropping Columns: Dropped the brand column with many missing values.
• Removing Missing Rows: Rows with missing values were dropped.

dataset.drop(columns=["brand"], inplace=True)
dataset.dropna(inplace=True)

2. Label Encoding

• Converted the product_category_tree into numerical labels to be used by the Logistic Regression model.

from sklearn.preprocessing import LabelEncoder
le = LabelEncoder()
dataset["en_product_category_tree"] = le.fit_transform(dataset["product_category_tree"])

3. TF-IDF Vectorization

• Used TfidfVectorizer to convert textual product descriptions into numerical vectors.

from sklearn.feature_extraction.text import TfidfVectorizer
tfidf = TfidfVectorizer()
X_tfidf = tfidf.fit_transform(dataset["description"])

4. Train-Test Split

• Split the data into 80% training and 20% testing sets.

from sklearn.model_selection import train_test_split
X = dataset["description"]
y = dataset["product_category_tree"]
x_train, x_test, y_train, y_test = train_test_split(X_tfidf, y, test_size=0.2, random_state=42)

5. Model Selection: Logistic Regression

• Trained a Logistic Regression model on the vectorized data.

from sklearn.linear_model import LogisticRegression
lr = LogisticRegression()
lr.fit(x_train, y_train)

6. Hyperparameter Tuning: GridSearchCV

• Tuned the model’s hyperparameters (C and solver) using grid search.

from sklearn.model_selection import GridSearchCV
parameters = {'C': [0.1, 1, 10, 100], 'solver': ['newton-cg', 'lbfgs', 'liblinear']}
grid_search = GridSearchCV(LogisticRegression(), parameters, cv=5)
grid_search.fit(x_train, y_train)

7. Evaluation

• Used the test dataset to evaluate the model’s performance using accuracy, classification report, and confusion matrix.

from sklearn.metrics import accuracy_score, classification_report, confusion_matrix

# Predictions on test data
y_pred = best_model.predict(x_test)

# Test accuracy
test_accuracy = accuracy_score(y_test, y_pred)

# Classification report and confusion matrix
class_report = classification_report(y_test, y_pred)
conf_matrix = confusion_matrix(y_test, y_pred)

8. Save Predictions

• Saved predictions for unseen test data into a CSV file.

test_data['predictions'] = predictions
test_data.to_csv('test_data_with_predictions.csv', index=False)

---

Results

Test Accuracy

The Logistic Regression model achieved a Test Accuracy of 98.29% on the test dataset.

Classification Report

The classification report shows precision, recall, and F1-score for each product category. Here’s the breakdown of the report:

Category	Precision	Recall	F1-score	Support
Automotive	0.98	0.99	0.99	202
Baby Care	0.94	0.89	0.91	36
Bags, Wallets & Belts	1.00	0.95	0.97	41
Clothing	1.00	1.00	1.00	1051
Computers	0.96	0.97	0.96	95
Footwear	1.00	1.00	1.00	206
Home Decor & Festive Needs	0.98	0.97	0.97	126
Jewellery	1.00	1.00	1.00	668
Kitchen & Dining	0.95	0.99	0.97	112
Mobiles & Accessories	0.98	0.98	0.98	150
Pens & Stationery	0.91	0.71	0.79	68
Tools & Hardware	1.00	0.98	0.99	66
Toys & School Supplies	0.70	0.85	0.76	46
Watches	1.00	1.00	1.00	120
Overall accuracy			0.98	2987

Confusion Matrix

The confusion matrix provides a visual breakdown of how well the model performs across different categories. It shows the number of true positive, false positive, true negative, and false negative predictions for each category.

Here is a heatmap visualization of the confusion matrix:

In this confusion matrix:

• True Labels are represented along the y-axis.
• Predicted Labels are represented along the x-axis.
• Diagonal elements represent correct predictions (true positives).
• Off-diagonal elements represent misclassifications (false positives and false negatives).

---

Conclusion

The model performed exceptionally well, achieving a high accuracy and excellent precision, recall, and F1-scores across most categories. Categories like Clothing, Jewellery, and Watches saw perfect classification results, while categories like Toys & School Supplies had slightly lower performance.

---

Future Work

Possible enhancements include:

• Trying different machine learning algorithms such as Random Forest or XGBoost.
• Implementing data augmentation to handle categories with fewer samples and improve overall performance.
• Deploying the model in production as an API to classify products in real-time.

---