DQ4781/LF-ShapAnalysis

An Analysis of Lassa Fever Outbreaks in Nigeria using Machine Learning Models and Shapley Values Decoding the Drivers of Lassa Fever Outbreaks in Nigeria Insights from Random Forest and XGBoost Models with SHAP Analysis ## Overview 📋 This project uses machine learning models and Shapley value analysis to **identify which ecoclimatic features have the most significant impact on Lassa Fever outbreaks in Nigeria**. The project includes code for data preprocessing, exploratory data analysis, feature selection, model training, and evaluation. ## Features & Dataset 📀 Using Google Datasets Search, the following keywords were used to narrow the search for a suitable dataset: "Lassa Fever", "Nigeria", "Lassa Fever Nigeria", "Lassa", and "Lassa Infections". With this combination of keywords, Google Datasets Search yielded a dataset consisting of weekly epidemilogical updates of Lassa Fever infections from the Nigerian CDC from 2012 to 2019. Also included were various feature classes such as: climate, land cover data, population data, and health facilities covariates. The dataset can be viewed here. ## ML Model 🌳 ### Preprocessing The following steps were taken to preprocess the data for both models: - Read dataset from CSV into pandas DataFrame - Create separate DataFrames to test predictive ability on data from 2018 & 2019 - Encode categorical variables into their respective numerical representations - Define the feature matrix by selecting all columns except the target variable - Split the feature matrix into training and testing sets using the 'train_test_split' function from scikit-learn ### Training & Hyperparameter Tuning In order to find the best model, the following steps were taken: - Define a parameter grid using the 'param_grid' dictionary to specify hyperparameters and their corresponding values for tuning - Consider hyperparameters such as 'max_depth', 'n_estimators', and 'learning_rate' that control tree depth, boosting rounds, and step size shrinkage - Use RandomizedSearchCV to perform a randomized search over the defined hyperparameter grid - Call …