Reducing Execution Time of Pixel-Based Machine Learning Classification Algorithms Using Parallel Processing Concept

Parallel processing is essential in machine learning to meet the computational requirements resulting from the complexity of algorithms and the size of the dataset, by taking advantage of the computational resources of parallel processing that can distribute computational operations across multiple processors. Which contributes to significant improvements in performance and time efficiency. This research demonstrated the impact of parallel processing on the performance and time efficiency of machine learning for pixel-based image classification techniques. The methodology includes pre-processing the Oxford IIIT Pet dataset, from which 4 cat images were selected. The performance of two supervised machine learning classifiers, decision tree, and random forest (10, 100, 500, and 1000 trees) were compared and implemented in two ways with and without parallel processing. The data is split in two ways: the first is by splitting the data by 70% for training data and 30% for testing data and the second is by cross-validation by splitting the data into four folds. The research aims to compare the accuracy and timely scales of machine learning models with and without parallel processing. The results showed a strong predictive power of the algorithms with an accuracy of 97.5%, while the training times were significantly reduced in parallel from 88.83 to 15.88 seconds for the RF100 model for image no. 2. This reflects the effectiveness of parallel processing in improving the performance of machine-learning models for pixel-based image classification. The proposed system was programmed using MATLAB 2021 language tools.