Matt's Portfolio

20-30 second overview of each project. Use this to navigate through all my projects.

Predicting strength changes in the powerlifting population over time using ARIMA models

Read the report here

View code at the original repo here

• Using arima forecasting, I predict change in the powerlifting population over time with a root Mean Squared Error (root MSE) of ~ 6.14 IPF-points, which can be interpreted as approximately 6kg difference in total weight lifted across all three lifts (for a 185 pound male).
• The following is visual comparison between a few tentative ARIMA models and the ground truth of unseen data:
• After selecting two of the best models, I forecasted one year into the future (waiting on the results from the data source to be updated to validate this):
• This data uses the powerlifting competition scores from hundreds of thousands of powerlifters.

Identifying lawns from Satellite Images of Properties Using Mask R-CNN

Read the report here

View the python notebook and instructions for running the model here

• Identified lawns with a mean Average Precision (mAP) of 86.02%.
• Adopted Matterport Inc's implementation of Mask R-CNN as our implementation.
• used image augmentation to overcome sample size barriers (only 85 original images).

Using Reisistance Training Data to Facilitate Strength Progression.

Read the project report here

View the repository with code here

• I captured over a year of my weight training data in hopes of increasing my squat, bench press, and deadlift strength.
• As a result of the analysis, I put 30 pounds on my bench press (310-340 lbs) , 50 pounds on my squat (405-455) , and about 65 pounds on my deadlift (430-495) after nearly a year of stagnation.
• The process was as follows: I followed an 8-10 week program (also called a block), analyzed the results of the training block, updated a few parameters of it (volume per week, average intensity per week, week to week variability in intensity, and week to week variability in volume), then followed the updated version for 8-10 more weeks. There were a total of four blocks performed. Here are the results:
• As you can see, block 1 was an experimental block without much results, but the insights from block 1 lead to stellar increases in strength in blocks 2 and 3. Block 4 was another experimental block like block 1, and insights were taken from it that did lead to training improvement after cessation of data collection.
• No formal statistical modeling was used for the following reasons: (1) I'm not estimating population parameters; (2) time series forecast is useless, as I am knowingly going to manipulate causal factors that will change the process governing that forecast; (3) formal experimental design is untennable.