The hardware and bandwidth for this mirror is donated by METANET, the Webhosting and Full Service-Cloud Provider.
If you wish to report a bug, or if you are interested in having us mirror your free-software or open-source project, please feel free to contact us at mirror[@]metanet.ch.
Full Documentation & Tutorials: https://gonrui.github.io/prepkit/
“When Z-Score fails, use M-Score.”
prepkit is a comprehensive R package
designed for the prepkitocessing of longitudinal behavioral data, with a
specific focus on gerontology, digital health, and sensor
analytics.
Its flagship feature is the M-Score (Mode-Range Normalization), a novel algorithm designed to detect anomalies in data characterized by “habitual plateaus” (e.g., daily step counts, heart rate), where traditional methods like Z-Score or Min-Max scaling often fail due to skewed distributions and high-frequency routine noise.
prepkit is rigorously tested on Linux, macOS,
and Windows, with compatibility verified up to R 4.5
(development version).
You can install the stable version from GitHub:
# install.packages("devtools")
devtools::install_github("Gonrui/prepkit")| Function | Algorithm / Description | Use Case |
|---|---|---|
norm_mode_range |
M-Score (New!) | Detects frailty/falls in elderly behavioral data by suppressing routine noise. |
trans_boxcox |
Robust Box-Cox | MLE-optimized power transform. Auto-handles non-positive values. |
trans_yeojohnson |
Yeo-Johnson | Power transform natively supporting negative values. |
norm_l2 |
Spatial Sign | Projects data onto a unit hypersphere (L2 Norm). Ideal for high-dim clustering. |
pp_plot |
Density Visualizer | Instant “Before vs. After” visualization for normality checks. |
Real-world sensor data often contains routine plateaus (e.g., an older adult consistently walking 3000 steps) and sensor noise (floating-point jitter).
M-Score handles both elegantly:
library(prepkit)
# 1. Simulate Sensor Data
# - Routine: ~3000 steps (with sensor jitter like 3000.1, 2999.9)
# - Anomaly: 200 steps (Fall/Frailty)
# - Anomaly: 6000 steps (Hyperactivity)
steps <- c(3000.1, 3000.2, 2999.8, 200.0, 3000.0, 6000.0)
# 2. Apply M-Score (with precision protection)
# The 'digits' parameter rounds values internally to identify the semantic mode,
# while preserving the precision of the anomalies.
m_scores <- norm_mode_range(steps, digits = 0)
# 3. View Results
print(data.frame(Raw = steps, M_Score = m_scores))The M-Score transforms data based on its Mode Interval (the routine plateau). Unlike Z-Score which penalizes stability (low variance), M-Score treats the most frequent range as the “Safe Zone” (Score = 0).
The transformation function \(M(x)\) is defined as:
\[ M(x) = \begin{cases} -\frac{k_L - x}{k_L - k_{min}} & \text{if } x < k_L \quad (\text{Left Tail / Frailty}) \\ 0 & \text{if } k_L \le x \le k_R \quad (\text{Routine Plateau}) \\ \frac{x - k_R}{k_{max} - k_R} & \text{if } x > k_R \quad (\text{Right Tail / Hyper}) \end{cases} \]
Validated via symbolic computation (Wolfram Mathematica) for strict monotonicity.
If you use prepkit or the M-Score algorithm in your
research, please cite:
Gong, R. (2026). M-Score: A Robust Normalization Method for Detecting Anomalies in Longitudinal Behavioral Data. arXiv preprint.
MIT © Rui Gong (Tokyo Metropolitan Institute of Gerontology)
These binaries (installable software) and packages are in development.
They may not be fully stable and should be used with caution. We make no claims about them.