📊 statistics

A Novel Three-Parameter Extended Weibull Distribution for Health Data Modelling

This paper proposes a novel three-parameter extended Weibull distribution that offers superior flexibility for modeling heavily tailed health data, demonstrating its effectiveness through comprehensive statistical derivations, simulation studies, and a comparative analysis on a fracture dataset where it outperforms five competing models.

Original authors: Isqeel Ogunsola, Nurudeen Ajadi, Gboyega Adepoju

Published 2026-04-07

📖 4 min read☕ Coffee break read

CC BY 4.0

Original authors: Isqeel Ogunsola, Nurudeen Ajadi, Gboyega Adepoju

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine you are a doctor trying to predict how long a specific medical device (like a hip replacement or a heart valve) will last before it breaks. To do this, you need a mathematical "ruler" to measure time and failure.

For decades, statisticians have used a ruler called the Weibull Distribution. It's a workhorse in the medical and engineering world. Think of it as a standard, reliable tape measure. It works great for most things.

But here's the problem: Real life is messy. Sometimes, medical data has "extreme" outliers—devices that last much longer than expected, or fail in weird, unpredictable ways. The standard Weibull tape measure is too rigid. It can't stretch or shrink to fit these weird shapes. It's like trying to wrap a square peg in a round hole; the fit is poor, especially at the edges (the "tails" of the data).

The Solution: A New, Flexible Ruler

The authors of this paper, Isqeel Ogunsola and his team, decided to build a new, super-flexible ruler. They took the standard Weibull tape measure and added a special "stretchy" mechanism to it.

They call their new invention the SMPtW Distribution (SMP-transformed Weibull).

Here is how they did it, using a simple analogy:

1. The "Shape-Shifting" Magic (The SMP Method)

Imagine the standard Weibull distribution is a lump of clay. It has a basic shape, but it's a bit stiff.
The authors used a technique called the SMP method (named after the initials of the researchers who invented the technique). Think of SMP as a magical pair of hands that can knead the clay.

It adds a third knob (a new parameter called $\lambda$ ) to the ruler.
Turning this knob allows the ruler to stretch, compress, or twist the data curve.
If you turn the knob to a specific setting, the new ruler magically turns back into the old, standard Weibull ruler. This means the new one is a "super-version" that includes the old one as a special case.

2. Why Do We Need This? (The "Tail" Problem)

In statistics, the "tail" of a graph represents the extreme events—the very long-lasting devices or the very early failures.

The Old Ruler: If you have a patient who survives 50 years when the average is 10, the old ruler says, "That's impossible," or "That's a huge mistake." It can't model that extreme well.
The New Ruler: The SMPtW ruler is flexible enough to say, "Ah, yes, that extreme event is possible." It bends to fit the data, even when the data is "heavily tailed" (full of extreme outliers).

What Did They Do in the Paper?

The authors didn't just invent the ruler; they put it through a rigorous gym workout to prove it works:

The Blueprint (Math Properties): They wrote down the exact mathematical formulas for everything: the average wait time, the probability of failure, and how "spread out" the data is. They proved the math holds up.
The Simulation (The Test Drive): They used computers to generate thousands of fake data sets. They tried to guess the settings of their new ruler using these fake sets.
- Result: As they fed more data into the computer, their guesses got incredibly accurate. The ruler learned the true shape of the data very well.
The Real-World Test (The Fracture Data): This is the most important part. They took real data about 76 broken knee implants (fracture data).
- They tried to fit the standard Weibull ruler.
- They tried five other "upgraded" rulers from other scientists.
- They tried their new SMPtW ruler.

The Verdict: The SMPtW ruler was the clear winner. It fit the broken knee data better than any of the others. It had the lowest "error score" (AIC) and the highest "confidence score" (P-value).

The Takeaway

Think of this paper as the introduction of a Swiss Army Knife to a world that only had standard pocket knives.

Old Way: Use a standard tool. It works okay for simple jobs, but fails when the job gets weird or extreme.
New Way: Use the SMPtW tool. It has an extra lever that lets it adapt to complex, messy, real-world health data.

Why does this matter?
In medicine, getting the math right matters. If you underestimate how long a device lasts, you might replace it too early (wasting money). If you overestimate it, a patient might suffer a sudden failure. This new, flexible distribution helps doctors and engineers make more accurate predictions, especially for those tricky, extreme cases that usually break the math.

In short: The authors found a way to make a classic statistical tool more flexible, tested it thoroughly, and proved it's the best tool yet for modeling complex health data.

1. Problem Statement

The Weibull distribution is a cornerstone in reliability engineering, medicine, and health sciences for modeling time-to-event data. However, the standard two-parameter Weibull distribution suffers from significant limitations:

Lack of Tail Flexibility: It often fails to provide a satisfactory fit for extreme events or data with heavy tails.
Inability to Model Complex Hazard Rates: It cannot adequately capture non-monotonic hazard rates (e.g., bathtub-shaped or unimodal failure rates) often found in complex biological and health datasets.
Inadequacy for Real-World Data: No single distribution fits all scenarios, necessitating the development of more flexible extensions that maintain mathematical tractability.

2. Methodology

The authors propose a novel three-parameter extension of the Weibull distribution, termed the SMPtW (SMP Transformed Weibull) distribution, utilizing the SMP transformation method (named after authors Rasool et al., 2025).

Transformation Mechanism: The SMP method transforms a base cumulative distribution function (CDF), $G(y)$ , into a new CDF, $F(y; \lambda)$ , using a transformation parameter $\lambda$ :
$F(y; \lambda) = \frac{e^{(\log \lambda)\bar{G}(y)} - \lambda}{1 - \lambda}$
where $\bar{G}(y) = 1 - G(y)$ is the survival function of the base distribution.
Derivation: By applying this transformation to the standard Weibull CDF, the authors derived the PDF and CDF of the SMPtW distribution.
Statistical Properties: The paper provides rigorous mathematical derivations for:
- Reliability measures (Survival and Hazard functions).
- Quantile function and Median.
- Moments (raw and central), Moment Generating Function (MGF), and Characteristic Function.
- Stress-strength reliability ( $R = P(Y_1 > Y_2)$ ).
- Order statistics, Mean Residual Life, and Mean Waiting Time.
- R´enyi Entropy.
- Mode (derived numerically).
Estimation and Simulation:
- Parameter Estimation: Maximum Likelihood Estimation (MLE) was used. The log-likelihood function was derived, and the resulting non-linear equations were solved numerically (using methods like Newton-Raphson).
- Random Number Generation: An inverse transform approach was implemented to generate random variates for simulation.
- Simulation Study: A Monte Carlo simulation was conducted with sample sizes ranging from 50 to 1,000 to evaluate the bias and Mean Squared Error (MSE) of the MLEs.

3. Key Contributions

Novel Distribution: Introduction of the SMPtW distribution, which adds a third parameter ( $\lambda$ ) to the standard Weibull, enhancing flexibility without sacrificing tractability.
Sub-models: The distribution generalizes several existing models:
- Reduces to the standard Weibull when $\lambda = 1$ .
- Reduces to the Exponential distribution when $\lambda = 1, \phi = 1$ .
- Reduces to the Rayleigh distribution under specific parameter constraints.
Comprehensive Theoretical Framework: The paper fills a gap in the literature by providing a complete set of statistical properties for the SMP-transformed Weibull, including entropy and stress-strength reliability, which were previously unexplored for this specific combination.
Methodological Review: The paper updates the review of SMP-transformed distributions, highlighting that while other base distributions (Lomax, Pareto, Kumaraswamy) have been transformed, the Weibull application was missing.

4. Results

Simulation Performance:
- The MLEs for the shape ( $\phi$ ) and scale ( $\beta$ ) parameters showed consistent convergence, with bias and MSE decreasing as sample size increased.
- The transformation parameter ( $\lambda$ ) exhibited higher bias and variability in small-to-moderate samples, converging more slowly. This suggests potential numerical challenges in estimating $\lambda$ with limited data.
Real Data Application (Health Data):
- Dataset: The model was applied to a dataset of 76 Kelvar 373/epoxy fatigue fracture times (often used as a proxy for health/lifetime data in reliability studies).
- Comparison: SMPtW was compared against five competitors: Standard Weibull, Exponentiated Weibull, Transmuted Weibull, Sine Alpha Power Weibull, and the base Weibull.
- Performance Metrics: SMPtW achieved the best fit across all criteria:
  - AIC: 247.34 (lowest among all models).
  - BIC, AICc, HQIC: All ranked first (lowest values).
  - Kolmogorov-Smirnov (K-S) Statistic: 0.0881 (lowest deviation).
  - P-value: 0.5668 (highest, indicating the null hypothesis of a good fit cannot be rejected).
- Visual Fit: Histograms and CDF plots confirmed that SMPtW captured the data's tail behavior and central tendency better than competing models.

5. Significance and Conclusion

Superiority in Health Modeling: The study demonstrates that the SMPtW distribution is a superior alternative for modeling heavily tailed health and reliability data, particularly where extreme events are of interest.
Flexibility: The transformation mechanism allows the distribution to adapt to non-monotonic hazard rates and complex tail structures that standard Weibull extensions miss.
Future Directions: The authors acknowledge that while the model performs well, the estimation of the transformation parameter $\lambda$ can be numerically challenging. They recommend future research into robust estimation techniques and Bayesian approaches to address these numerical issues and improve convergence in small sample sizes.

In summary, this paper successfully extends the Weibull family using the SMP transformation, providing a mathematically rigorous and empirically superior tool for analyzing complex health and reliability data.