Health Predict

This dataset is originally from the National Institute of Diabetes and Digestive and Kidney Diseases. The objective of the dataset is to diagnostically predict whether or not a patient has diabetes, based on certain diagnostic measurements included in the dataset. Several constraints were placed on the selection of these instances from a larger database. In particular, all patients here are females at least 21 years old of Pima Indian heritage.

https://www.kaggle.com/datasets/uciml/pima-indians-diabetes-database

Technical Note: The prediction will be made by two algorithms: Random Forest and Logistic Regression. Here are descriptions of these [algorithms:

Random Forest: Random Forest is a machine learning algorithm used for classification and regression tasks. It works by creating multiple decision trees during training and then combines their results to make a final prediction. Each tree in the forest gives a prediction, and the algorithm takes the majority vote (in classification) or the average (in regression) to decide the final output.

Random Forest helps reduce the risk of overfitting and improves accuracy by using a collection of trees rather than relying on a single decision tree.

Logistic Regression: Logistic Regression is a machine learning algorithm used for binary classification, where the goal is to predict one of two possible outcomes (like yes/no, 0/1). It works by estimating the probability that a given input belongs to a particular class.

Instead of giving a direct prediction, Logistic Regression uses a mathematical function called the logistic (or sigmoid) function to output a probability between 0 and 1. If the probability is above a certain threshold (usually 0.5), it classifies the data as one class (e.g., 1), and if it’s below that threshold, it classifies it as the other class (e.g., 0).

It’s called “regression” because it predicts probabilities, but it’s mainly used for classification tasks.

Result Description: The approach taken in this project is that each time, 10 data points are randomly selected as test data, and the remaining data are considered as training data. Then, the accuracy of both algorithms is recorded. This process is repeated 10 times, and the final accuracy shown is the average of these 10 runs for each algorithm. The result is that, for the data in this problem, both algorithms perform almost the same. Of course, the program has been run multiple times to reach this conclusion.

Some of Results: Random Forest: 0.89 Logistic Regression: 0.9099999999999999

Random Forest: 0.75 Logistic Regression: 0.72

Random Forest: 0.76 Logistic Regression: 0.77

Random Forest: 0.8400000000000001 Logistic Regression: 0.8300000000000001

Git repo of this project: https://github.com/SevenSkyConsulting/health-predict

Predicting Diabetes: A Data-Driven Approach

Diabetes is a growing health concern worldwide, and early detection is key to managing it effectively. This project set out to predict whether an individual is likely to develop diabetes based on a range of diagnostic measurements. With the help of machine learning, we aimed to create a model that can analyze these medical indicators and make predictions that might one day support healthcare professionals in early diagnosis.

Data Behind the Prediction The data used in this project comes from the National Institute of Diabetes and Digestive and Kidney Diseases. The dataset includes information from Pima Indian women over the age of 21, providing insights into several health metrics. Each entry contains parameters like:

Number of pregnancies Glucose levels Blood pressure Skin thickness Insulin levels BMI (Body Mass Index) Diabetes pedigree function (a measure of hereditary diabetes risk) Age

With these attributes, we aimed to predict a binary outcome: the presence or absence of diabetes. This dataset, curated for diagnostic purposes, offers a focused view, helping us create a model that could potentially aid in diabetes screening.

The Methods: Random Forest and Logistic Regression To tackle this binary classification problem, we tested two algorithms: Random Forest and Logistic Regression. Each has its strengths and unique approach to making predictions.

Random Forest Imagine a forest of decision trees, each trained on slightly different subsets of data. Each tree makes its own prediction, and the forest as a whole combines these predictions for a final decision. By averaging the results (for regression) or taking a majority vote (for classification), Random Forest reduces the chance of overfitting, often delivering robust, accurate predictions. For example, in our project, if 80% of the trees predict diabetes, the forest will classify the individual as likely to develop diabetes.

Logistic Regression Logistic Regression is simpler yet highly effective for binary classification. Rather than predicting a direct yes or no, it calculates the probability of an individual developing diabetes. Using a sigmoid (or logistic) function, it outputs a value between 0 and 1. If this probability crosses a threshold—typically 0.5—the model predicts “diabetes present”; otherwise, it predicts “no diabetes.” Logistic Regression is particularly useful for understanding the impact of each feature on the outcome.

Process and Results: Testing for Robustness To evaluate the performance of our models, we used a cross-validation approach:

Each time, we randomly selected 10 data points as a test set, leaving the rest as training data. We recorded the accuracy of both algorithms for each test set. This process was repeated 10 times to ensure variability, and we averaged the results to get a reliable measure of accuracy. In our results, both models performed similarly:

Random Forest achieved an average accuracy that hovered around 75% to 89% across different runs. Logistic Regression also performed consistently, with accuracies around 72% to 91%. While Logistic Regression slightly outperformed Random Forest on average, the two algorithms demonstrated comparable results, indicating that either could be a viable choice for this problem.

Key Insights and Future Considerations This analysis highlights the feasibility of using simple, accessible data points to make impactful health predictions. Although both models performed well, a few future enhancements could strengthen the predictions:

Additional Data and Features: Incorporating more health parameters or lifestyle factors could provide deeper insights and improve accuracy. Feature Importance Analysis: Understanding which features (like glucose levels or BMI) have the most impact on the prediction could help in fine-tuning the model and focusing on critical health markers. Broader Patient Demographics: Extending the dataset to include individuals from various backgrounds would make the model more generalizable and valuable across a broader population.

Conclusion By combining patient data and machine learning, this project offers a glimpse into how predictive modeling can support medical decisions. Though not a substitute for professional diagnosis, a model like this could serve as an early screening tool, potentially catching diabetes in its early stages and enabling timely intervention.