Optimization of Random Forest Hyperparameters with Genetic Algorithm in Classification of Lung Cancer

Lung cancer is a type of cancer with the highest death rate compared to other cancers. Cancer can be classified using histopathological methods which are obtained using biopsies. Manual classification of cancer on histopathological images is work intensive and highly susceptible to human error. Cancer classification from histopathological images can be done using computer assistance using computer vision and machine learning. This research proposes the following stages: data collection, feature extraction from images, feature selection, building a Random Forest model, optimizing hyperparameters using a Genetic Algorithm, and evaluating the performance of the model. The histopathological images that have been collected will have their color and texture features extracted. The extraction process produces 9 RGB features and 9 HSV features for color features. Meanwhile, texture features produce 6 types of features, namely dissimilarity, correlation, homogeneity, contrast, ASM, and energy, which are then searched for values from four different angles to produce 24 texture features. A total of 42 features were produced. All these features are then selected using the correlation coefficient and the remaining 24 features will be used to build a classification model using Random Forest. The classification model that has been built is then optimized by setting hyperparameters automatically so that the resulting model is reliable and better than general models. The hyperparameters that are optimized are
$n$
estimators, max depth, max features, and criterion. By using a Genetic Algorithm, all hyperparameters are adjusted automatically to get hyperparameters with the best model performance. The Random Forest model with hyperparameters with default values succeeded in getting an accuracy of 98.82 % and a 10-fold cross-validation value of 99.39%. Meanwhile, the model that has been optimized using the Genetic Algorithm with the best hyperparameters
$n$
estimators = 300, max depth = 100, max features = log2, and criterion = entropy produces an accuracy of 98.83% and a 10-fold cross-validation value of 99.50%. The Random Forest model with hyperparameters optimized using the Genetic Algorithm succeeded in outperforming the Random Forest model with default hyperparameters. It is proven that optimizing hyperparameters using a Genetic Algorithm can improve the performance of the Random Forest model.