Using a DNN Model to Predict Weight Loss on a Ketogenic Diet

---

Alexandre Gomes, a tech enthusiast and blogger, recently shared his experience using a deep neural network (DNN) model to predict and visualize his weight loss progress while following a ketogenic diet. Over the past 8 weeks, he lost over 20 kg and decided to leverage machine learning to better understand and project his weight loss journey.

Technical Overview

Gomes used a simple feedforward DNN model to capture the non-linear nature of his weight loss data. This approach is particularly useful for time series data where the relationship between inputs (time) and outputs (weight) can be complex. Here’s a breakdown of the key steps:

• Data Collection: Weekly weight measurements over 8 weeks.
• Model Selection: A feedforward DNN with multiple hidden layers to model non-linear relationships.
• Training: The model was trained to fit the collected data, minimizing the mean squared error (MSE) between predicted and actual weights.
• Inference: Once trained, the model can predict future weight loss based on the current trend.

Implementation Details

Data Preparation

• Input Features: Time in weeks.
• Output Labels: Weight in kilograms.
• Normalization: Normalized both input and output data to improve training stability and speed.

Model Architecture

• Layers:
  • Input layer: 1 neuron (time in weeks).
  • Hidden layers: 2 hidden layers with 32 neurons each, using ReLU activation functions.
  • Output layer: 1 neuron (predicted weight), using a linear activation function.
• Loss Function: Mean Squared Error (MSE) to measure the difference between predicted and actual weights.
• Optimizer: Adam optimizer with a learning rate of 0.001.

Training Process

• Dataset: 8 data points (weeks).
• Epochs: Trained for 500 epochs to ensure convergence.
• Batch Size: Used the entire dataset in each batch due to the small size.

Visualization

• Weight Loss Graph: Plotted actual and predicted weights over time.
• Calorie Deficit Graph: Calculated daily calorie needs using the Harris-Benedict Equation and plotted the difference from maintenance levels.

Key Insights

• Non-linear Trends: The DNN model effectively captured the non-linear trends in weight loss, providing more accurate predictions compared to linear models.
• Calorie Tracking: By comparing the predicted weight loss with the Harris-Benedict Equation, Gomes could estimate his daily calorie deficit and adjust his diet accordingly.

Harris-Benedict Equation

The Harris-Benedict Equation is a well-established formula used to estimate Basal Metabolic Rate (BMR) and Total Daily Energy Expenditure (TDEE). BMR is the number of calories your body needs to perform basic physiological functions at rest. TDEE includes all activities and can be adjusted for weight loss or gain.

• Formula:
  • For men: ( \text{BMR} = 88.362 + (13.397 \times \text{weight in kg}) + (4.799 \times \text{height in cm}) - (5.677 \times \text{age in years}) )
  • For women: ( \text{BMR} = 447.593 + (9.247 \times \text{weight in kg}) + (3.098 \times \text{height in cm}) - (4.330 \times \text{age in years}) )

Gomes used these calculations to create a graph showing his daily calorie needs and compared it with his actual weight loss rate.

Conclusion

By combining machine learning with traditional nutritional science, Gomes was able to gain deeper insights into his weight loss journey. The DNN model provided a powerful tool for visualizing and predicting future progress, while the Harris-Benedict Equation helped in understanding the caloric impact of his diet changes. This approach can be valuable for anyone looking to manage their weight more effectively using data-driven methods.