Below are a few of the tools I have created for the R Statistical Programming Language:
The Transition Pairing Method
This is a method for assessing the accuracy of change point detection (CPD) algorithms. It uses an extended Gale-Shapley algorithm to address the complex problem of defining true/false positives in the CPD setting. You can find the method in my PAutilities package (see GitHub for the development version). You can also learn more about the code (and how to implement it) in the associated vignette. The method is fully explained in my 2020 paper.
Two-Regression Energy Expenditure Models
These are 15 models that predict energy expenditure through a sophisticated pipeline. The models are for ActiGraph GT9X monitors worn on the hip, left wrist, right wrist, left ankle, or right ankle. For each of those locations, there are three models (accelerometer only, accelerometer+gyroscope, and accelerometer+gyroscope+magnetometer). The pipeline starts by classifying each data point as sedentary behavior, ambulation, or lifestyle activity. Energy expenditure is then predicted by means that differ depending on the prior classification (a constant value for sedentary behavior, a linear regression equation for ambulation, and a cubic regression equation for lifestyle activities). Like the Transition Pairing Method, you can find this method in an R package (with a development version on GitHub), along with a vignette that walks through the code, and a paper that has full detail.
This is a pattern recognition-based approach to looking at patterns of sedentary behavior (e.g., accumulation in short versus long bouts). We used data from the National Health and Nutrition Examination Survey to perform a clustering analysis that identified three profiles, which we called the Interrupted, Intermediate, and Prolonged profiles. The purpose of this was to address statistical challenges in temporal analysis, as well as to increase the interpretability and actionability of the variables. The paper was recently published, and you can view all of the code that went into it (including for the sample analysis of cardiovascular disease risk) on GitHub. You can also implement it through the R package (GitHub only), and there is a PDF vignette (view or download) to help with coding.
These models predict physical activity outcomes (energy expenditure for adults and activity intensity for youth) using a form of change point detection combined with neural networks and hand-built decision trees. The first models were published by Lyden et al. and Ellingson et al., before I introduced the youth-specific models and assembled everything into an R package (development version on GitHub).
The intake-balance method is a way of inferring energy intake from observed measurements of energy expenditure and body composition over time. I am working to promote open-source methods in this vein. Much of the work is still in progress, but you can read about some of it here.