===###Overview:This repository provides high-utility synthetic survival datasets generated using the CK4Gen framework, optimised to retain critical clinical characteristics for use in research and educational settings. Each dataset is based on a carefully curated ground truth dataset, processed with standardised variable definitions and analytical approaches, ensuring a consistent baseline for survival analysis.###===###Description:The repository includes synthetic versions of four widely utilised and publicly accessible survival analysis datasets, each anchored in foundational studies and aligned with established ground truth variations to support robust clinical research and training.#---GBSG2: Based on Schumacher et al. [1]. The study evaluated the effects of hormonal treatment and chemotherapy duration in node-positive breast cancer patients, tracking recurrence-free and overall survival among 686 women over a median of 5 years. Our synthetic version is derived from a variation of the GBSG2 dataset available in the lifelines package [2], formatted to match the descriptions in Sauerbrei et al. [3], which we treat as the ground truth.ACTG320: Based on Hammer et al. [4]. The study investigates the impact of adding the protease inhibitor indinavir to a standard two-drug regimen for HIV-1 treatment. The original clinical trial involved 1,151 patients with prior zidovudine exposure and low CD4 cell counts, tracking outcomes over a median follow-up of 38 weeks. Our synthetic dataset is derived from a variation of the ACTG320 dataset available in the sksurv package [5], which we treat as the ground truth dataset.WHAS500: Based on Goldberg et al. [6]. The study follows 500 patients to investigate survival rates following acute myocardial infarction (MI), capturing a range of factors influencing MI incidence and outcomes. Our synthetic data replicates a ground truth variation from the sksurv package, which we treat as the ground truth dataset.FLChain: Based on Dispenzieri et al. [7]. The study assesses the prognostic relevance of serum immunoglobulin free light chains (FLCs) for overall survival in a large cohort of 15,859 participants. Our synthetic version is based on a variation available in the sksurv package, which we treat as the ground truth dataset.###===###Notes:Please find an in-depth discussion on these datasets, as well as their generation process, in the link below, to our paper:https://arxiv.org/abs/2410.16872Kuo, et al. "CK4Gen: A Knowledge Distillation Framework for Generating High-Utility Synthetic Survival Datasets in Healthcare." arXiv preprint arXiv:2410.16872 (2024).###===###References:[1]: Schumacher, et al. “Randomized 2 x 2 trial evaluating hormonal treatment and the duration of chemotherapy in node-positive breast cancer patients. German breast cancer study group.”, Journal of Clinical Oncology, 1994.[2]: Davidson-Pilon “lifelines: Survival Analysis in Python”, Journal of Open Source Software, 2019.[3]: Sauerbrei, et al. “Modelling the effects of standard prognostic factors in node-positive breast cancer”, British Journal of Cancer, 1999.[4]: Hammer, et al. “A controlled trial of two nucleoside analogues plus indinavir in persons with human immunodeficiency virus infection and cd4 cell counts of 200 per cubic millimeter or less”, New England Journal of Medicine, 1997.[5]: Pölsterl “scikit-survival: A library for time-to-event analysis built on top of scikit-learn”, Journal of Machine Learning Research, 2020.[6]: Goldberg, et al. “Incidence and case fatality rates of acute myocardial infarction (1975–1984): the Worcester heart attack study”, American Heart Journal, 1988.[7]: Dispenzieri, et al. “Use of nonclonal serum immunoglobulin free light chains to predict overall survival in the general population”, in Mayo Clinic Proceedings, 2012.

E-commerce Sales Prediction Dataset

This repository contains a comprehensive and clean dataset for predicting e-commerce sales, tailored for data scientists, machine learning enthusiasts, and researchers. The dataset is crafted to analyze sales trends, optimize pricing strategies, and develop predictive models for sales forecasting.

📂 Dataset Overview

The dataset includes 1,000 records across the following features:

📊 Data Summary

General Properties

Date: - Range: 01-01-2023 to 12-31-2023. - Contains 1,000 unique values without missing data.

Product_Category: - Categories: Electronics (21%), Sports (21%), Other (58%). - Most common category: Electronics (21%).

Price: - Range: From 244 to 999. - Mean: 505, Standard Deviation: 290. - Most common price range: 14.59 - 113.07.

Discount: - Range: From 0.01% to 49.92%. - Mean: 24.9%, Standard Deviation: 14.4%. - Most common discount range: 0.01 - 5.00%.

Customer_Segment: - Segments: Regular (35%), Occasional (34%), Other (31%). - Most common segment: Regular.

Marketing_Spend: - Range: From 2.41k to 10k. - Mean: 4.91k, Standard Deviation: 2.84k.

Units_Sold: - Range: From 5 to 57. - Mean: 29.6, Standard Deviation: 7.26. - Most common range: 24 - 34 units sold.

📈 Data Visualizations

The dataset is suitable for creating the following visualizations: - 1. Price Distribution: Histogram to show the spread of prices. - 2. Discount Distribution: Histogram to analyze promotional offers. - 3. Marketing Spend Distribution: Histogram to understand marketing investment patterns. - 4. Customer Segment Distribution: Bar plot of customer segments. - 5. Price vs Units Sold: Scatter plot to show pricing effects on sales. - 6. Discount vs Units Sold: Scatter plot to explore the impact of discounts. - 7. Marketing Spend vs Units Sold: Scatter plot for marketing effectiveness. - 8. Correlation Heatmap: Identify relationships between features. - 9. Pairplot: Visualize pairwise feature interactions.

💡 How the Data Was Created

The dataset is synthetically generated to mimic realistic e-commerce sales trends. Below are the steps taken for data generation:

1. Feature Engineering:

   • Identified key attributes such as product category, price, discount, and marketing spend, typically observed in e-commerce data.
   • Generated dependent features like units sold based on logical relationships.

2. Data Simulation:

   • Python Libraries: Used NumPy and Pandas to generate and distribute values.
   • Statistical Modeling: Ensured feature distributions aligned with real-world sales data patterns.

3. Validation:

   • Verified data consistency with no missing or invalid values.
   • Ensured logical correlations (e.g., higher discounts → increased units sold).

Note: The dataset is synthetic and not sourced from any real-world e-commerce platform.

🛠 Example Usage: Sales Prediction Model

Here’s an example of building a predictive model using Linear Regression:

Written in python

import pandas as pd
from sklearn.model_selection import train_test_split
from sklearn.linear_model import LinearRegression
from sklearn.metrics import mean_squared_error, r2_score

# Load the dataset
df = pd.read_csv('ecommerce_sales.csv')

# Feature selection
X = df[['Price', 'Discount', 'Marketing_Spend']]
y = df['Units_Sold']

# Train-test split
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2, random_state=42)

# Model training
model = LinearRegression()
model.fit(X_train, y_train)

# Predictions
y_pred = model.predict(X_test)

# Evaluation
mse = mean_squared_error(y_test, y_pred)
r2 = r2_score(y_test, y_pred)

print(f'Mean Squared Error: {mse:.2f}')
print(f'R-squared: {r2:.2f}')