Skip to content

Latest commit

 

History

History
85 lines (65 loc) · 4.27 KB

Readme.md

File metadata and controls

85 lines (65 loc) · 4.27 KB
Raw

Size Chart Prediction Project

Overview

This project aims to predict accurate size charts for apparel sellers using machine learning. The model is trained on a dataset of body measurements and predicts appropriate clothing sizes.

Directory Structure

  • data/: Contains raw and processed datasets.
  • notebooks/: Jupyter notebooks for EDA and model development.
  • scripts/: Python scripts for data preprocessing, clustering, training, and predictions.
  • models/: Saved models.
  • tests/: Unit tests for scripts.
  • .vscode/: VSCode configuration files.
  • requirements.txt: List of dependencies.
  • README.md: Project overview and documentation.

Getting Started

  1. Clone the repository.
  2. Install dependencies: 
    pip install -r requirements.txt

Size Chart Prediction - Clustering Analysis

Overview

This project involves clustering analysis on a dataset of body measurements to identify distinct groups or clusters that could be used for creating personalized size charts. The primary goal is to leverage Agglomerative Clustering to segment the data into meaningful clusters.

Methods Used

  • Data Preprocessing: The dataset was cleaned by handling missing values, where random valid values or the mode were used to replace NaNs.
  • Clustering Algorithm: Agglomerative Clustering with 5 clusters was chosen as the method to segment the data.
  • Statistical Analysis: Post-clustering, statistical analysis was conducted to summarize each cluster's characteristics.
  • Visualization: Pair plots and cluster distributions were visualized to interpret the clustering results.

Key Statistics

Cluster Counts

The dataset was divided into 5 clusters with the following distribution:

Cluster Count
0 193
1 248
2 308
3 118
4 133

Cluster Centroids

Each cluster's centroid was calculated to understand the typical characteristics of the data points within that cluster. Below is a summary of key features:

Feature Cluster 0 Cluster 1 Cluster 2 Cluster 3 Cluster 4
Weight 84.25 59.08 72.71 102.26 53.02
Bust/Chest 37.98 37.13 40.00 39.54 37.68
Waist 31.29 28.73 37.95 34.87 37.57
Hips 35.85 32.82 43.30 39.50 43.37

(Include more features as required)

Cluster Standard Deviations

To understand the variability within each cluster, standard deviations were calculated for each feature:

Feature Cluster 0 Cluster 1 Cluster 2 Cluster 3 Cluster 4
Weight 6.07 6.89 6.21 6.51 6.21
Bust/Chest 6.22 5.91 6.73 6.55 5.24
Waist 5.88 4.18 5.10 6.75 4.11
Hips 6.55 4.83 4.64 6.74 3.91

(Include more features as required)

Interpretation of Results

  • Cluster 0: Represents a group with relatively higher weight and bust/chest measurements.
  • Cluster 1: Represents individuals with lower weight and average bust/chest measurements.
  • Cluster 2: Consists of individuals with balanced measurements across all features.
  • Cluster 3: Represents individuals with the highest weight and relatively higher bust/chest measurements.
  • Cluster 4: Represents a group with the lowest weight and higher waist and hips measurements.

These clusters can be used to create personalized size charts that cater to different body shapes and sizes.

Next Steps

  • Refinement: Consider experimenting with different numbers of clusters or clustering algorithms to see if better segmentation can be achieved.
  • Application: Implement these findings into a production environment for creating personalized size charts or other related applications.

Conclusion

The clustering analysis provided insights into distinct groups within the dataset, which can be valuable for personalizing size charts and improving customer satisfaction. Further refinement and validation could enhance the results.