anegrill/IDSN544_Chronic_Disease_Predictor

Chronic Disease Prediction Model

Problem Statement

Chronic diseases, such as diabetes, hypertension, and cardiovascular disorders, are heavily influenced by a combination of genetic, lifestyle, and social factors. The aim of this project is to develop a predictive model that can forecast the likelihood of individuals developing chronic diseases using social determinants of health and demographic information. By identifying at-risk populations, healthcare providers can focus preventive interventions and allocate resources more effectively.

Chronic Disease Overview

Chronic diseases tend to occur in older adults and can usually be controlled but not cured. The most common types of chronic disease are:

• —Cancer
• —Heart disease
• —Stroke
• —Diabetes
• —Arthritis

Objective

Primary Goal

• —Predict the likelihood of an individual developing chronic diseases using social, demographic, health history, and lifestyle factors.

Secondary Goal

• —Identify key social determinants (e.g., income, education, urban-rural residence) that are most predictive of chronic disease development.

Data Overview

The National Health Interview Survey (NHIS) has monitored the health of the nation since 1957. NHIS data on a broad range of health topics are collected through personal household interviews. Survey results have been instrumental in providing data to track health status, health care access, and progress toward achieving national health objectives.

Data Collection

The data and documentation used in this analysis were obtained from the National Health Interview Survey (NHIS) 2023, provided by the National Center for Health Statistics (NCHS).

Data Preprocessing

1. 1.Missing Values: Data with imputed Don't Know, Unknown, and Not Ascertained values; all features imputed except the following, which only have one value (binary may be imputed in workfile):
2. 2.HHRESPSA_FLG | Sample Adult is the household respondent or the proxy who lives in the… | 1 = Yes
3. 3.MAFLG_A | Medicaid reassignment flag | 1 = Reassigned to Medicaid from private
4. 4.PRFLG_A | Private reassignment flag | 1 = Reassigned to private from public
5. 5.CHFLG_A | CHIP reassignment flag | 1 = Reassigned to CHIP from…
6. 6.OPFLG_A | State-sponsored reassignment flag | 1 = Reassigned to other public from…
7. 7.OGFLG_A | Other government reassignment flag | 1 = Reassigned to other government from…
8. 8.Feature Engineering: Selected and using a combination of Chi-Squared, RF Importance, Lasso L1 Regularization, and Recursive Feature Elimination

Data Splitting

• —Training Set: 80% of the data.
• —Test Set: 20% of the data.

Data Quality and Limitations

• —Biases: Some features rely on self-reporting, which may introduce biases.
• —Missing Data: Small percentage of missing values were imputed.
• —Feature Limitations: The dataset may not capture all aspects of chronic disease risk as it is based on survey responses.

Data Storage and Access

The dataset is available in the data/ folder of this repository. If the data is not included, please refer to the Kaggle Dataset (https://www.kaggle.com/datasets/357bd3945441ce586b26eb45480a08c5a36b8c129fed11305e2bfe8958c6c7b9?select=test_full.csv) for access.

Data Usage and Licensing

National Center for Health Statistics, National Health Interview Survey, 2023. While the data are sourced from NCHS, the analyses, interpretations, and conclusions presented in this work are solely those of the author(s) and do not represent the views or positions of NCHS. NCHS is responsible only for the initial data collection and not for any subsequent interpretations or analyses derived from these data.

Model Details

A Random Forest Classifier was trained. Feature Engineering was completed through Chi2, RFI, L1 Lasso, and RFE. Resampling was performed using RandomOverSampling and Hyperparameters were tuned (nestimators, maxdepth, maxfeatures, minsamples_leaf, and bootstrap). Model was trained using NHIS 2023 data. Evaluation Metrics prioritized log loss and ROC-AUC, but includes accuracy, precision, recall, and F1 score.