Filip Aleksic

Filip - DATA 1150 Final Presentation

Filip Akelsic

Faculty Collaborator: Marissa Gray

About:

Filip had been collaborating with Dr. Marrisa Gray to develop an innovative course in the ENGN department, focused on wearable sensor data. One key component of the course involved students constructing simple wearable sensors to gather physiological data. However, the primary goal of the project was to integrate data science and machine learning into the curriculum, enabling students to analyze the data collected from their sensors, combined with demographic data they provided.

Specifically, the course aimed to predict whether a student was resting, walking, or running based on their heart rate and demographic information. Filip utilized Jupyter Notebooks to guide students through the entire lifecycle of a data science project. This comprehensive approach included data collection, data cleaning, data exploration, data visualization, and predictive modeling using various Python packages. By doing so, Filip ensured that students gained hands-on experience with both the practical and analytical aspects of data science.

transcript

Project Goal

Teach students how to analyze sensor data.

Data Collection

Health data is hard to find.
Sources:
- Apple Watch and Fitbit Data from Kaggle

Data Cleaning

Tools Used: Pandas & NumPy packages in Python
Processes:
- Dealing with structural errors
- Identifying and dealing with missing values
- Identifying and dealing with duplicates

Exploratory Data Analysis (EDA)

Tools Used: Matplotlib and Seaborn packages in Python
Types of Analysis:
- Univariate analysis
- Bivariate analysis
- Numerical vs. Categorical data
Goal: Expose students to different charts and graphs

Modeling

Goal: Predict whether a student is resting, walking, or running
Inputs:
- Sensor data: heart rate, steps, distance covered
- Demographic data: height, weight, age, gender
Target Variable: Activity
Algorithms:
- Non-linear classifiers: K-Nearest Neighbors (KNN), decision trees, random forests

Why Non-Linear Classifiers?

Effective for complex data patterns and interactions.

Model Evaluation

Metrics Used:
- Confusion Matrix
- Precision
- Recall

Personal Growth

Data Science:
- Proficiency with Jupyter Notebooks
- Domain knowledge
- Feature engineering
- Acquired various skills from other fellows
Teaching and Learning:
- Presenting technical findings
- Creating lectures and assignments
- Gathering resources
- Being a curious learner
Consulting:
- Project management
- Communication skills
- Presenting

Page updated

Report abuse