Tomertg/SmartFit_Dataset

SmartFit AI - Synthetic Fitness Dataset

Dataset Overview

This repository contains a synthetic dataset generated for the SmartFit AI project. The dataset consists of 10,000 user profiles designed to train a machine learning model for personalized workout recommendations. It simulates various biological and preference-based factors including age, gender, weight, height, fitness goals, equipment availability, and injuries.

The data is stored in Parquet format (Hugging Face native) for high efficiency and fast loading.

Repository Contents

• —Copy of Final Project.ipynb (4): Colab's notebook
• —fitness_embeddings.npy (2): Pre-computed vector embeddings of the user profiles (generated via Sentence Transformer).
• —imp88.png: Visualization of the Exploratory Data Analysis (EDA).
• —imp888.png: Visualization of the Embeddings Analysis (PCA Clustering).

Exploratory Data Analysis (EDA)

The visualization below demonstrates the distribution and characteristics of the synthetic data. This analysis ensures the dataset is balanced across different categories and logically consistent regarding injuries and recommended plans.

[image]

Embeddings and User Segmentation

To analyze user similarity and validate the data quality, we utilized a Sentence Transformer model to convert textual user profiles into high-dimensional vectors. The visualization below displays these embeddings reduced to 2D space using PCA. The distinct clusters indicate that the system successfully differentiates between different user types and their corresponding workout requirements.

[image]