callmekofi / cactus Goto Github PK
View Code? Open in Web Editor NEWThis project forked from mikbjorn/cactus
An image classification problem identifying the presence or absence of cactus in a 32 by 32 pixel image. The 17,500-image database was split 60% train, 20% validation, 20% test. Parallel processing was implemented for image processing and model building. Feature engineering included the creation edge features using the difference between adjacent pixels. Support vector machine, random forest, Xgboost, and convolutional neural network (CNN) models were fit. The CNN fit with TensorFlow and Keras outperformed all other models with 97.5% classification accuracy. By using a weighted average of the Xgboost, Random Forest, and CNN models a classification accuracy of 97.7% was achieved. This was a team project. I was responsible for the image processing, feature engineering, Random Forest, Xgboost, and weighted average models.