Comprehensive Analysis of Smartphone Sensor Data Using Machine Learning Techniques

Introduction

This project provides a comprehensive analysis of smartphone sensor data collected from volunteers performing various activities. The analysis encompasses:

• Preprocessing steps such as outlier reduction and dimensionality reduction using Principal Component Analysis (PCA).
• Clustering analysis performed using K-means and Density-Based Spatial Clustering of Applications with Noise (DBSCAN).

The objective is to uncover meaningful patterns and structures in the data to gain insights into human activity recognition through smartphone sensors.

---

Dataset Explanation

The dataset consists of sensor readings captured from smartphones worn by 30 volunteers aged 19-48 while performing six activities:

• WALKING
• WALKING_UPSTAIRS
• WALKING_DOWNSTAIRS
• SITTING
• STANDING
• LAYING

Data Collection

• Device: Samsung Galaxy S II positioned on the waist.
• Sensors: 3-axial linear acceleration and 3-axial angular velocity at 50Hz.
• Preprocessing:
  • Signals were preprocessed to separate gravitational and body motion components.
  • Noise filters and fixed-width windows (2.56 seconds with 50% overlap, 128 readings per window) were applied.
  • A Butterworth low-pass filter (0.3 Hz cutoff frequency) separated gravitational forces from body motion components.

Feature Extraction

Feature vectors were extracted from each window by calculating variables from the time and frequency domains, representing aspects of human activity like acceleration, velocity, and frequency components.

---

Data Preprocessing

Outlier Reduction

• Identified and removed using the z-score method to improve data quality and reduce noise.

Dimensionality Reduction with PCA

• PCA was used to:
  • Reduce the number of features while preserving data variance.
  • Address multicollinearity.
  • Improve computational efficiency.
• The cumulative explained variance ratio determined the optimal number of principal components to retain.

---

Clustering Analysis

K-means Clustering

• Method: Applied to PCA-transformed data to identify distinct clusters.
• Evaluation:
  • Elbow Method and Silhouette Score determined the optimal number of clusters.
  • Cluster centroids analyzed for relevance to human activities.

DBSCAN Clustering

• Method: Clustered the preprocessed data based on density.
• Parameters: Optimal eps and min_samples determined using the k-distance plot and Silhouette Score.
• Results:
  • Identified clusters of varying shapes and sizes.
  • Highlighted potential noise points for further investigation.

---

Results and Discussion

K-means Clustering Results

• Produced clusters based on data variance.
• Optimal number of clusters: 3.
• Cluster characteristics:
  • Cluster 0: Activities involving high linear acceleration (e.g., walking, walking upstairs, walking downstairs).
  • Cluster 1: Stationary activities (e.g., sitting, standing).
  • Cluster 2: Complex movements indicating transitions between postures or activities.

DBSCAN Clustering Results

• Identified clusters based on density.
• Highlights:
  • Effective in detecting clusters with irregular shapes or varying densities.
  • Identified 499 noise points, indicating potential outliers or anomalous behaviors.

Outlier Reduction

• Improved the quality and robustness of clustering results by mitigating noise impact using the z-score method.

---

Conclusion

The comprehensive analysis of smartphone sensor data using PCA, K-means, and DBSCAN yielded valuable insights into human activity recognition. Future work could explore:

• Advanced feature engineering techniques.
• Refinement of clustering algorithms.
• Validation using additional datasets.

---

How to Run the Notebook

1. Dependencies: Ensure the following Python libraries are installed:

   numpy
   pandas
   matplotlib
   seaborn
   sklearn
   

2. Run the Notebook: Open and execute the cells sequentially in a Jupyter Notebook environment.
3. Input Data: Ensure the dataset is located in the specified directory within the notebook.

---

Acknowledgments

This project utilizes smartphone sensor data collected from volunteers for research purposes in human activity recognition.