Week 6 and 7 updates

 

Blog Update 1

Week 6 – 31/03/2025

Blog Creation and Backup

  • The project blog was officially created to serve as a central repository for progress tracking, updates, and backup on GitHub.

  • Relevant files were uploaded for safekeeping and version control, establishing a workflow for consistent documentation and data management.

Week 7 – 24/04/2025

Part 1

Feature Engineering and Initial Exploration

The analysis began with the selection and visualization of five engineered features. Using MATLAB, basic exploratory techniques such as box plots and scatter plots were employed to gain an initial understanding of data distributions and the relationships between predictors and the target variable.

Principal Component Analysis (PCA)

PCA was applied to both unnormalized and normalized data to assess the variance captured by each principal component. Pareto charts were used to visualize component significance, while 2D and 3D scatter plots facilitated the identification of patterns and potential clustering in reduced-dimensional space. The intrinsic dimensionality of the dataset was estimated using dy/dx plots, indicating a meaningful dimensionality of approximately nine.







Correlation Analysis

A heatmap of the normalized dataset was generated to investigate linear correlations between variables. This revealed several features with minimal or no correlation to glucose levels and highlighted collinearity among some variables, suggesting that dimensionality reduction and feature selection would be beneficial.

Curvilinear Component Analysis (CCA)

Given the apparent non-linearity in the dataset, Curvilinear Component Analysis was performed. Compared to PCA, CCA provided more informative and balanced low-dimensional representations. The dy/dx plots post-CCA transformation further validated the earlier dimensionality estimates and indicated better-preserved interclass distances.



Preliminary Model Testing

MATLAB’s Regression Learner App was used to prototype a narrow neural network model. Initial results demonstrated a root mean square error (RMSE) of approximately 40.7 mg/dL. Visual analysis suggested that this value was influenced by outliers; excluding them could potentially reduce the RMSE to below 20 mg/dL. However, the model exhibited a low R² value (-2.06), indicating that further model refinement is necessary.

Key Findings

  • Outlier Removal: Boxplots effectively identified outliers, which were removed to improve the reliability of statistical measures. Exceptionally high glucose values (~220 mg/dL) were retained, as they reflect clinically plausible cases.

  • Feature Relevance: Features such as std_ch3, crestf_ch2, Higher-order statistical moments were found to have weak correlations with the target variable, making them potential candidates for exclusion.

  • Dimensionality Insight: PCA and CCA confirmed that a small subset of components (approximately nine) captured the majority of useful variance, offering significant opportunities for dimensionality reduction.

  • Model Behavior: Despite low R² values, the predicted outputs demonstrated clustering in physiologically relevant glucose ranges (90–130 mg/dL), indicating some level of underlying structure that may be captured more effectively with advanced modeling techniques.

Future Work

Following the insights gained during this stage, the next phase of the project will involve:

  • Refinement of feature selection to eliminate noisy or redundant predictors.

  • Exploration of more sophisticated models, including deep learning architectures and ensemble methods.

  • Enhanced model validation through cross-validation and hyperparameter tuning.

Part 2

GlucoData Analysis and Predictive Modeling

This project presents a systematic approach to analyzing data obtained from a glucometer prototype. The dataset includes measurements and lifestyle information from 461 patients, compiled to support predictive modeling for glucose levels.

Dataset Overview

The initial dataset comprises 21 features, including demographic and physiological variables such as:

  • Age

  • Gender

  • Height (cm)

  • Weight (kg)

  • Body Mass Index (BMI)

  • Heart Rate (bpm)

  • Blood Glucose Level (mmol/L)

These features were extracted from the raw data to construct a refined 6-feature dataset focused on lifestyle and health indicators relevant to glucose levels.

Data Preprocessing

To enhance data quality and model performance, the following preprocessing steps were applied:

  • Removal of missing values and outliers, resulting in a reduced dataset of 318 valid entries.

  • Normalization using Z-score transformation, which standardizes features to have a mean of 0 and a standard deviation of 1.

Data visualization techniques, including box plots and scatter plots, were used to assess the distribution and detect anomalies in both raw and normalized datasets.

Dimensionality Reduction

To reduce data complexity while preserving critical information:

  • Principal Component Analysis (PCA) was performed, reducing the dataset from 6 to 4 dimensions.

  • Visual tools such as Pareto charts and biplots provided insights into the relationships between features.

  • Curvilinear Component Analysis (CCA) was also employed for further dimensionality reduction and visualization, highlighting potential clusters and outliers.

Correlation Analysis

A correlation matrix heatmap of the normalized data was used to identify redundancy and interdependencies among features. Notably, BMI and weight exhibited a strong correlation, indicating potential for feature reduction in future modeling efforts.

Machine Learning and Model Evaluation

The Regression Learner App in MATLAB was utilized to train and evaluate various regression models:

  • The Ensemble Regression model yielded the best performance, though with modest results:

    • Training R² = 0.06

    • Testing R² = 0.12

Despite limited predictive accuracy, the model serves as a foundational benchmark for future improvements.

Performance Visualization

  • Predicted vs. Actual plots revealed significant variance, indicating low predictive precision.

  • A response plot illustrated substantial prediction errors, likely due to non-linear relationships and weak feature-response correlations.

Future Work

To improve the predictive performance of the model, future efforts will focus on:

  • Advanced feature engineering

  • Hyperparameter optimization

  • Implementation of non-linear and deep learning models

  • Expansion of the dataset, if possible, to enhance training efficacy

Comments

Post a Comment

Popular posts from this blog

EE461 Project Breakdown

  EE461 Project Blog: Glucose Estimation Model Development Welcome to the official blog for our EE461 project! This space will serve as a central hub for logging our progress, sharing insights, backing up key developments, and tracking milestones throughout the project. Posts will be updated fortnightly, ensuring consistent documentation of our journey as we work towards building a robust glucose estimation model. Project Overview The primary goal of this project is to develop a reliable machine learning model capable of estimating glucose levels based on available physiological or sensor data. This task involves building the model itself and ensuring the quality and relevance of the input data, engineering meaningful features, and validating model performance through sound evaluation metrics. Project Objectives To achieve our goals, the project will be guided by the following key objectives: Data Preprocessing Cleaning and preparing the raw dataset is a crucial first ste...
Read more

Week 9 to 11 updates

Image
  Blog Update 2 Weeks 9 to 11 – 28/04/2025 to 16/05/2025 Feature Selection and Early Model Testing Feature Diagnostics and Ranking Using MATLAB’s Feature Diagnostic Designer , we calculated, ranked, and extracted the most relevant features for glucose estimation. These features were then applied across various quick-to-train machine learning models to measure performance improvements. The best-performing model achieved an R² value of 0.14 , marking a 14% improvement over the previously implemented neural network model. This result was obtained using voltage sensor data , reinforcing the value of sensor-based inputs in glucose prediction. ANN Prototyping and Evaluation An Artificial Neural Network (ANN) was designed and tested using lifestyle data . The model achieved an accuracy of 55% on both validation and test sets. While modest, these results demonstrate potential for model development, especially in scenarios where only lifestyle indicators are available. As...
Read more

Week 12 and 13 updates

  Blog Update 3 Week 12 and 13 20/05/25 - 03/06/25 Inference Pipeline Development and Final Presentation Preparation This week marked the transition from model experimentation to a fully functional inference pipeline, laying the groundwork for real-time glucose prediction. Alongside technical implementation, we also finalized the presentation materials and prepared a demo package for showcasing the end-to-end system. Inference Pipeline Implementation To simulate real-world prediction scenarios, a test sample was extracted and pushed to the end of the dataset. This allowed us to mimic unseen data inference while preserving a ground truth value for validation. The complete dataset, including this test sample, was then passed through the full preprocessing workflow, which included normalization and dimensionality reduction. Curvilinear Component Analysis (CCA) was used to transform the feature space, reducing it to 45 dimensions. This technique, already proven effective in prio...
Read more