Thermal-Anomaly-Classifiers

End-to-end machine learning pipeline capable of analyzing thermal images and classifying object types of hotspot • Conducted comprehensive exploratory data analysis (EDA), including missing value imputation, categorical encoding, feature scaling, and correlation analysis. Generated and displayed a correlation matrix to highlight key predictors • Implemented five classifiers—Decision Trees, SVM, Naive Bayes, KNN, and Random Forests—evaluating performance based on accuracy, precision, recall, F1-score, and confusion matrices • Executed hyperparameter tuning using Grid Search on 3–5 critical parameters per model and implemented SMOTE to handle class imbalances, achieving an average 15% improvement in F1-score and accuracy across all classifiers, with the top model (Random Forests) achieving a 90% score in all evaluation metrics