Python, R & Machine Learning Academy

Completed pair plot analysis revealing strong linear relationships between key medical features and disease progression. BMI and S5 (LDL) show the clearest positive correlations, while other features demonstrate more complex patterns. The diagonal KDE plots highlight normal-like distributions suitable for linear regression modeling. These visual insights guided our feature selection for Bayesian MCMC analysis.

• Built Bayesian regression model predicting crop yields from 8 soil parameters
• Achieved R² = 0.466, identifying nitrogen as primary positive driver
• Discovered potassium deficiency (30 vs 40-50 kg/ha for N/P)
• Generated automated dashboard with posterior distributions & uncertainty intervals
• Quantified variable importance with 95% HDI for risk-based decision making

Spatial Disease Risk Prediction Dashboard 📊🗺️

I’m excited to share a key component of my latest project focused on spatial modeling and risk prediction using advanced statistical and data science techniques.

This dashboard provides a comprehensive view of disease risk across a geographic area:

🔹 Predicted Disease Risk Map
Visualizes spatial variation in disease risk using interpolation techniques. High-risk clusters are clearly identified, helping in targeted intervention planning.

🔹 Prediction Uncertainty
Highlights the uncertainty associated with predictions, which is crucial for understanding model confidence and improving decision-making.

🔹 High-Risk Zones (Top 25%)
Identifies and maps the most critical regions that require immediate attention, enabling efficient resource allocation.

🔹 Risk Distribution Profile
Shows the statistical distribution of predicted risk values, including mean and threshold levels for low and high-risk classification.

This project integrates spatial statistics, data visualization, and predictive modeling to support data-driven healthcare strategies.

📊 Nutrient Response Curves with Bayesian Uncertainty

As part of my project in Bayesian modeling and agricultural data analysis, I explored how key soil nutrients influence crop yield while accounting for uncertainty.

🔍 The plots show the relationship between yield (tons/ha) and four important factors:

* Nitrogen
* Phosphorus
* Potassium
* Organic Matter

Each subplot includes:
🔹 Scatter points – observed data
🔹 Blue line – mean predicted effect
🔹 Shaded region – 95% credible interval (uncertainty)

📈 Key Insights:

* Nitrogen & Phosphorus: Both show a slight positive effect on yield, but the relationship is relatively weak, suggesting diminishing returns at higher levels.
* Potassium: Minimal impact observed, indicating it may not be a limiting factor in this dataset.
* Organic Matter: Shows a clearer positive trend, highlighting its importance in improving soil productivity and yield.

✅ Why this matters:
Using Bayesian methods allows us not only to estimate effects but also to quantify uncertainty, leading to more reliable and interpretable insights for decision-making in agriculture.

📊 Posterior Predictive Check – Model Validation

As part of my Bayesian modeling project, I performed a posterior predictive check to evaluate how well the model captures the underlying data distribution.

🔹 The blue histogram represents the observed data (actual yield values), while the **red shaded distribution shows the simulated data generated from the posterior predictions.

🔹 A strong overlap between observed and predicted distributions indicates that the model is successfully capturing the key patterns in the data.

🔹 In this case, the model performs well around the central region (≈ 2.5–3.5 tons/ha), suggesting accurate prediction of typical outcomes. Slight differences in the tails highlight areas where the model could be further refined.

✅ Key Insight
Posterior predictive checks are essential in Bayesian analysis to assess model fit and ensure that predictions are realistic and reliable.

This step strengthens confidence in the model before making decisions or drawing conclusions.