Random Forest Model for Predicting Grayscale Digits on Images

Abstract

The random forest is one of the fastest, powerful and easy to use techniques with the ability of performing regression and classification task recorded in the biblical study of machine learning. It grows into forest with some number of decision trees as a hybrid system. The problem of missing data which results from over-fitting can best be handled by random forest. The process places a very big barrier on how well the existing methods and functions. This also affects the success metrics in the process of training, testing and validating classifiers in order to produce a more generalized and accurate results for predicting grayscale digits on images. This model is usually employed in replacing the continuous variables with the median values while computing the correct weighted proximity average of all the missing data values. Our aim is to develop an efficienct random forest model using the classification and regressor classifiers with some highquality of input data by introducing weighted average and the best in majority vote. This model has been trained and tested using the MNIST dataset with some adjusted hyper-parameter values which defined and improved the general performance of the combined trees in the forest. The simulation was done using python programming language and resulted to 99% and 90% metrics of accuracy in comparison with the random forest classification and regression approach respectively.