📊 Body Performance Analytics

Production-Grade Data Cleaning, Exploratory Analysis, & ML Pipeline

An end-to-end Machine Learning data preparation framework transforming raw multi-feature physical performance records into clean, engineered, and scaled model-ready datasets.

---

📌 Table of Contents

• 1. Project Overview & Abstract
• 2. Data Pipeline Architecture
• 3. 14-Step Physiological Data Cleaning Audit
• 4. Exploratory Data Analysis & Key Insights
• 5. Feature Engineering & Mathematical Specifications
• 6. Data Transformations (Encoding & Scaling)
• 7. Machine Learning Model Readiness
• 8. Repository Directory Structure
• 9. Installation & Quick Start Guide
• 10. Author & Acknowledgments

---

1. Project Overview & Abstract

Predictive performance tier classification (Class A: Elite, Class B: Good, Class C: Average, Class D: Below Average) plays a central role in sports science, military readiness evaluation, and public health risk modeling. However, physical test measurements collected under field conditions frequently suffer from human data-entry typos, sensor zeroing errors, physiological impossibilities, and non-linear feature interactions.

This repository provides a production-ready Data Science & Machine Learning Preprocessing Pipeline engineered on the Body Performance Dataset (13,393 participant records across 12 physical and physiological features).

🌟 Key Engineering Highlights:

• Domain-Informed Cleaning over Naive Trimming: Standard $1.5 \times \text{IQR}$ statistical truncation aggressively discards top-tier athletic outliers. Instead, we implement a 14-step sequential rule system based on physiological limits, eliminating corrupt records while preserving genuine performance signals.
• 7 Derived Physiological Features: Engineered domain metrics including Body Mass Index (BMI), Fat/Lean Body Mass, Mean Arterial Pressure (MAP), Pulse Pressure, Relative Grip Strength, and Power Index.
• Multiclass Balance & Model Readiness: Target class is perfectly balanced (~25% in each tier A–D), eliminating majority-class bias. Dual scaled exports (StandardScaler and MinMaxScaler) enable immediate consumption by distance-based, linear, and neural network algorithms.

---

2. Data Pipeline Architecture

┌────────────────────────────────────────────────────────────────────────────────────────┐
│                                1. DATA INGESTION & AUDIT                               │
│  - Input: raw `bodyPerformance.csv` (13,393 rows x 12 features)                        │
│  - Check schema, data types, missing values (0 missing), & duplicates (1 duplicate)    │
└───────────────────────────────────────────┬────────────────────────────────────────────┘
                                            │
                                            ▼
┌────────────────────────────────────────────────────────────────────────────────────────┐
│                     2. SEQUENTIAL PHYSIOLOGICAL CLEANING (14 RULES)                    │
│  - Step 01–05 : Blood pressure logic errors (diastolic >= systolic, 0s, hypertensive) │
│  - Step 06–08 : Anthropometric boundaries (height < 140cm, weight < 30kg, fat > 60%)  │
│  - Step 09–12 : Flexibility & BMI anomalies (sit-and-reach < -15cm, BMI < 17)        │
│  - Step 13–14 : Broad jump clipping (< 25cm) & Sit-ups contradiction (Class D error)   │
│  - Output: 13,278 clean rows (115 anomalous rows removed = 0.86% data loss)            │
└───────────────────────────────────────────┬────────────────────────────────────────────┘
                                            │
                                            ▼
┌────────────────────────────────────────────────────────────────────────────────────────┐
│                        3. EXPLORATORY DATA ANALYSIS (EDA)                              │
│  - Density distribution overlay validation (Raw vs. Cleaned distributions)             │
│  - Target tier balance audit & bivariate class boxplots                                │
│  - Gender dimorphism split violin plots & Pearson correlation heatmap                  │
│  - 5-Feature pairplot separation matrix                                                │
└───────────────────────────────────────────┬────────────────────────────────────────────┘
                                            │
                                            ▼
┌────────────────────────────────────────────────────────────────────────────────────────┐
│                           4. PHYSIOLOGICAL FEATURE ENGINEERING                         │
│  - Anthropometric: BMI, Fat Mass (kg), Lean Mass (kg)                                  │
│  - Cardiovascular: Mean Arterial Pressure (MAP), Pulse Pressure                        │
│  - Performance: Relative Grip Strength (Grip / Weight), Power Index (Sit-ups x Jump)   │
└───────────────────────────────────────────┬────────────────────────────────────────────┘
                                            │
                                            ▼
┌────────────────────────────────────────────────────────────────────────────────────────┐
│                        5. CATEGORICAL ENCODING & FEATURE SCALING                       │
│  - Binary Encoding : `gender` (Male = 1, Female = 0)                                   │
│  - Ordinal Encoding: `class`  (A = 3, B = 2, C = 1, D = 0)                              │
│  - Scaling Options : StandardScaler (Z-score) & MinMaxScaler (0-1)                     │
└───────────────────────────────────────────┬────────────────────────────────────────────┘
                                            │
                                            ▼
┌────────────────────────────────────────────────────────────────────────────────────────┐
│                            6. PRODUCTION CSV EXPORTS                                   │
│  - Output 1: `data_clean.csv`        (13,278 rows x 20 columns - Unscaled)            │
│  - Output 2: `data_preprocessed.csv` (13,278 rows x 19 columns - StandardScaled)      │
└────────────────────────────────────────────────────────────────────────────────────────┘

---

3. 14-Step Physiological Data Cleaning Audit

Note

Sequential execution is critical: subsequent rules (e.g., BMI calculation) depend on prior rules (height/weight validation) already being clean, preventing double-counting of overlapping anomalies.

Step	Target Feature	Rule Condition	Action Taken	Physiological Rationale	Impact / Count
01	diastolic, systolic	diastolic >= systolic	Drop Row	Diastolic pressure must strictly be lower than systolic pressure.	16 rows dropped
02	systolic	systolic == 0	Drop Row	Zero blood pressure reading represents sensor zeroing / missing error.	1 row dropped
03	diastolic	diastolic == 0	Drop Row	Zero blood pressure reading represents sensor zeroing / missing error.	1 row dropped
04	systolic	systolic > 190 mmHg	Drop Row	Medically extreme resting hypertension; unfit for performance testing.	3 rows dropped
05	diastolic	diastolic > 120 mmHg	Drop Row	Hypertensive crisis territory; unfit for performance testing.	1 row dropped
06	body fat_%	body fat_% > 60%	Drop Row	Body fat percentage above 60% is a clear data entry error for active subjects.	1 row dropped
07	height_cm	height_cm < 140 cm	Drop Row	Impossible adult height (population age range 21–64 years).	1 row dropped
08	weight_kg	weight_kg < 30 kg	Drop Row	Impossible adult body mass for tested age demographic.	1 row dropped
09	sit and bend forward_cm	flexibility > 45 cm	Drop Row	Exceeds anatomical limit of sit-and-reach flexibility box.	1 row dropped
10	sit and bend forward_cm	flexibility < -15 cm	Drop Row	Extreme negative scores represent severe tightness/injury or test error.	84 rows dropped
11	BMI_raw	BMI < 17	Drop Row	Severe underweight beyond adult physiological safety margins.	4 rows dropped
12	height_cm, weight_kg, age	Height $>165$cm & Weight $<40$kg & Age $&gt;20$	Drop Row	Cross-feature thinness sanity check secondary safety net.	0 rows dropped
13	broad jump_cm	broad jump_cm < 25 cm	Clip (25)	Jump recording typo; clipping preserves row signal without data loss.	10 rows clipped
14	sit-ups counts, class	sit-ups > 70 & class == 'D'	Drop Row	Logical contradiction: elite sit-up score assigned to worst class tier D.	1 row dropped

Cleaning Audit Summary:

• Initial Row Count: $13,393$
• Total Removed Rows: $115$
• Final Clean Row Count: $13,278$ ($99.14%$ retention, $0.86%$ data loss)

---

4. Exploratory Data Analysis & Key Insights

🎯 Top 7 Actionable Business & Physiological Findings

1. ⚖️ Perfect Target Balance:
   The target variable class contains an equal distribution across all 4 performance tiers ($\approx 25%$ per class), ensuring machine learning classifiers will not suffer from class imbalance bias.

2. 🏋️ Dynamic Strength & Power as Top Discriminators:
   sit-ups counts and broad jump_cm exhibit strong monotonic alignment with performance tiers. Participants in Class A average $> 53$ sit-ups and $> 225$ cm broad jump distances, whereas Class D averages $< 31$ sit-ups.

3. 🚺🚹 Gender Dimorphism & Relative Strength Normalization:
   Male participants display higher absolute gripForce ($46.8$ kg vs. $28.3$ kg) and broad jump distance. However, normalizing grip force by body weight (relative_grip = $\text{Grip} / \text{Weight}$) aligns relative athletic performance across genders.

4. 🧘 Non-Linear Flexibility Threshold:
   sit and bend forward_cm serves as a strict boundary metric. Negative flexibility scores ($< 0$ cm) are overwhelmingly concentrated in Class C and Class D, indicating posterior chain tightness.

5. 🫀 Cardiovascular Risk Stratification:
   Systolic and diastolic blood pressure show strong linear correlation ($r = 0.68$). Deriving Mean Arterial Pressure (MAP) ($\text{Diastolic} + \frac{1}{3}\Delta\text{BP}$) condenses blood pressure into a unified cardiovascular load index.

6. 🎯 Label Contradiction Resolution:
   Step 14 detected participant records scoring $> 70$ sit-ups assigned to Class D (worst tier). Eliminating label corruption prevents neural networks and decision trees from learning conflicting split boundaries.

7. 📊 Zero-Distortion Data Integrity:
   Overlaid density histograms confirm that the 14-step cleaning pipeline eliminated corrupt records while preserving the underlying continuous distributions of all valid features.

---

5. Feature Engineering & Mathematical Specifications

We derive 7 domain-specific features grounded in exercise physiology and cardiovascular metrics:

$$\text{BMI} = \frac{\text{Weight (kg)}}{\left(\frac{\text{Height (cm)}}{100}\right)^2}$$ $$\text{Fat Mass (kg)} = \text{Weight (kg)} \times \left(\frac{\text{Body Fat \%}}{100}\right)$$ $$\text{Lean Mass (kg)} = \text{Weight (kg)} - \text{Fat Mass (kg)}$$ $$\text{MAP (mmHg)} = \text{Diastolic} + \frac{1}{3} \times \left(\text{Systolic} - \text{Diastolic}\right)$$ $$\text{Pulse Pressure (mmHg)} = \text{Systolic} - \text{Diastolic}$$ $$\text{Relative Grip} = \frac{\text{Grip Force (kg)}}{\text{Weight (kg)}}$$ $$\text{Power Index} = \text{Sit-ups Count} \times \text{Broad Jump (cm)}$$

📋 Engineered Feature Definitions & Data Types

Feature Name	Derived Formula	Data Type	Physiological Concept
BMI	$\text{Weight} / \text{Height}^2$	float64	Overall body mass index relative to height.
fat_mass_kg	$\text{Weight} \times (\text{Body Fat %} / 100)$	float64	Total body fat mass in kilograms.
lean_mass_kg	$\text{Weight} - \text{Fat Mass}$	float64	Fat-free muscle and skeletal tissue mass.
MAP	$\text{Diastolic} + \frac{1}{3}(\text{Systolic} - \text{Diastolic})$	float64	Mean Arterial Pressure (average organ perfusion pressure).
pulse_pressure	$\text{Systolic} - \text{Diastolic}$	float64	Vascular stiffness and cardiac stroke effort indicator.
relative_grip	$\text{Grip Force} / \text{Weight}$	float64	Bodyweight-normalized upper body strength.
power_index	$\text{Sit-ups Count} \times \text{Broad Jump}$	float64	Combined muscular endurance and lower-body explosive power.

---

6. Data Transformations (Encoding & Scaling)

🔤 Categorical & Target Encoding

• Gender (gender): Binary Mapping ($\text{Male} \rightarrow 1$, $\text{Female} \rightarrow 0$).
• Performance Class (class): Ordinal Mapping ($\text{Class A} \rightarrow 3$, $\text{Class B} \rightarrow 2$, $\text{Class C} \rightarrow 1$, $\text{Class D} \rightarrow 0$).

🔢 Numerical Feature Scaling

We scale numerical predictor features using two complementary transformations:

1. StandardScaler (Z-Score Standardization): $$\mu = 0, \quad \sigma = 1 \quad \implies \quad z = \frac{x - \mu}{\sigma}$$ Recommended for: Logistic Regression, Support Vector Machines (SVM), K-Nearest Neighbors (KNN), and Neural Networks.

2. MinMaxScaler (0–1 Normalization): $$x_{\text{scaled}} = \frac{x - x_{\min}}{x_{\max} - x_{\min}}$$ Recommended for: Neural Networks with bounded activation functions, Distance Metrics, and Gradient-based Optimizers.

---

7. Machine Learning Model Readiness

Dataset Export	File Name	Rows	Columns	Scaling Transformation	Target Column	Primary Model Use-Case
Clean Unscaled	data_clean.csv	13,278	20	Unscaled (Raw Units)	class ($3, 2, 1, 0$)	Random Forest, XGBoost, LightGBM, Decision Trees
Preprocessed Scaled	data_preprocessed.csv	13,278	19	StandardScaler ($z$-score)	class ($3, 2, 1, 0$)	SVM, Logistic Regression, Neural Networks, KNN

---

8. Repository Directory Structure

.
├── 01_EDA_and_Preprocessing_Pipeline.ipynb   # Production Jupyter Notebook (Main Execution Pipeline)
├── README.md                                 # Recruiter-Ready GitHub Documentation
├── requirements.txt                          # Production Python Dependency List
├── bodyPerformance.csv                        # Raw Dataset (13,393 rows x 12 columns)
├── data_clean.csv                             # Cleaned Dataset with 7 Engineered Features (Unscaled)
├── data_preprocessed.csv                      # Fully Preprocessed & StandardScaled Dataset
├── images/
│   └── hayper_digi_dashboard.png              # Project Header Banner Image
├── assets/
│   └── banner.png                            # Backup High-Resolution Banner Image
└── 01,02. EDA (Data Analysis and Preprocessing)/
    ├── 01. EDA/
    │   └── Final/
    │       ├── 01_EDA _new.ipynb             # Exploratory Analysis Working Notebook
    │       └── 02_Preprocessing_(1).ipynb    # Preprocessing Rules Prototype Notebook
    └── Data/
        ├── bodyPerformance-original data.csv  # Original Source CSV Copy
        └── Data Cleaned.csv                   # Legacy Clean Dataset Copy

---

9. Installation & Quick Start Guide

1. Clone the Repository

git clone https://github.com/KarimMostafa/Body-Performance-Analytics.git
cd Body-Performance-Analytics

2. Create & Activate Virtual Environment

# Windows (PowerShell / CMD)
python -m venv venv
.\venv\Scripts\activate

# macOS / Linux
python3 -m venv venv
source venv/bin/activate

3. Install Required Dependencies

pip install -r requirements.txt

4. Run the Jupyter Notebook

jupyter notebook 01_EDA_and_Preprocessing_Pipeline.ipynb

---

10. Author & Acknowledgments

👨‍💻 Karim Mostafa

Data Analyst & Machine Learning Engineer

Primary Tech Stack & Tools:
Python • Pandas • NumPy • Matplotlib • Seaborn • Scikit-Learn • Jupyter • LaTeX

---

This project is licensed under the MIT License.

_{Built with ❤️ by Karim Mostafa — 2026}