Abstract:We introduce Longitudinal Predictive Conformal Inference (LPCI), a novel distribution-free conformal prediction algorithm for longitudinal data. Current conformal prediction approaches for time series data predominantly focus on the univariate setting, and thus lack cross-sectional coverage when applied individually to each time series in a longitudinal dataset. The current state-of-the-art for longitudinal data relies on creating infinitely-wide prediction intervals to guarantee both cross-sectional and asymptotic longitudinal coverage. The proposed LPCI method addresses this by ensuring that both longitudinal and cross-sectional coverages are guaranteed without resorting to infinitely wide intervals. In our approach, we model the residual data as a quantile fixed-effects regression problem, constructing prediction intervals with a trained quantile regressor. Our extensive experiments demonstrate that LPCI achieves valid cross-sectional coverage and outperforms existing benchmarks in terms of longitudinal coverage rates. Theoretically, we establish LPCI's asymptotic coverage guarantees for both dimensions, with finite-width intervals. The robust performance of LPCI in generating reliable prediction intervals for longitudinal data underscores its potential for broad applications, including in medicine, finance, and supply chain management.

What problem does this paper attempt to address?

This paper aims to solve the problem of constructing prediction intervals in longitudinal data. Specifically, existing conformance prediction methods, when dealing with time - series data, usually focus on univariate settings and lack the cross - sectional coverage ability when applied to each time series separately. For longitudinal data, in order to ensure cross - sectional and asymptotic longitudinal coverage, existing methods often need to create infinitely wide prediction intervals, which are not feasible in practical applications. Therefore, the paper proposes a new distribution - free conformance prediction algorithm - Longitudinal Prediction Conformance Inference (LPCI) to solve these problems. ### Main contributions: 1. **Solve the problems of existing methods**: The LPCI method can ensure both longitudinal and cross - sectional coverage without using infinitely wide intervals. 2. **Innovative method**: By modeling the residual data as a quantile fixed - effect regression problem and using the trained quantile regressor to construct prediction intervals, LPCI provides an effective solution. 3. **Theoretical guarantee**: The paper proves the asymptotic coverage guarantee of LPCI in the cross - sectional and longitudinal dimensions, and the width of the prediction interval is finite. 4. **Experimental proof**: Through extensive experiments, the paper shows that LPCI is superior to existing benchmark methods in terms of cross - sectional coverage and longitudinal coverage rate, and the width of its prediction interval has higher adaptability. ### Specific problem description: - **Cross - sectional coverage**: Refers to the coverage situation of all groups (such as patients, stocks or customers) at a given time point. - **Longitudinal coverage**: Refers to the coverage situation of future values on the time series of each group. - **Limitations of existing methods**: Most existing methods cannot provide cross - sectional coverage guarantee when applied to each time series; and in order to ensure longitudinal coverage, it is necessary to create infinitely wide prediction intervals, which are not feasible in practical applications. ### Solutions: - **LPCI method**: By modeling the residuals as a quantile fixed - effect regression problem and using the trained quantile regressor to construct prediction intervals, LPCI can ensure cross - sectional and longitudinal coverage without using infinitely wide intervals. - **Theoretical analysis**: The paper proves the asymptotic coverage guarantee of LPCI in the cross - sectional and longitudinal dimensions, and the width of the prediction interval is finite. - **Experimental verification**: Through experiments on two actual data sets (COVID and EEG), the paper shows the superior performance of LPCI. ### Experimental results: - **Cross - sectional coverage**: LPCI performs well in cross - sectional coverage and reaches the expected coverage rate. - **Longitudinal coverage**: LPCI is superior to existing benchmark methods in longitudinal coverage rate. - **Adaptability of interval width**: The width of the prediction interval of LPCI has higher adaptability and can dynamically adjust the interval width according to the uncertainty of the model. In conclusion, by proposing the LPCI method, this paper solves the key problem of constructing prediction intervals in longitudinal data, provides theoretical and empirical support, and shows its potential in practical applications.

Conformal Predictions for Longitudinal Data

Conformal Prediction Intervals with Temporal Dependence

Conformal Prediction Regions for Time Series using Linear Complementarity Programming

Posterior Conformal Prediction

Sequential Predictive Conformal Inference for Time Series

Efficient and minimal length parametric conformal prediction regions

Conformal Prediction with Learned Features

Beyond Conformal Predictors: Adaptive Conformal Inference with Confidence Predictors

A Conformal Prediction Approach to Explore Functional Data

Adaptive Conformal Prediction Intervals Using Data-Dependent Weights With Application to Seismic Response Prediction

Conformal prediction for multi-dimensional time series by ellipsoidal sets

Conformal Prediction Under Covariate Shift

Conformal Prediction for Causal Effects of Continuous Treatments

Conformalization of Sparse Generalized Linear Models

Conformal Prediction Intervals for Remaining Useful Lifetime Estimation

Conformal Prediction with Temporal Quantile Adjustments

conformalInference.multi and conformalInference.fd: Twin Packages for Conformal Prediction

Multi-Source Conformal Inference Under Distribution Shift

Spatial Conformal Inference through Localized Quantile Regression

Length Optimization in Conformal Prediction