Word Sense Disambiguation is a generic task that need to be solved in Natural Language Processing, especially for languages with highly complex morphological structure, such as Arabic. The goal of this study is to implement a novel approach for the machine learning-based algorithm intended to increase the level of the Arabic disambiguation performance by using the context-aware strategy. The proposed algorithm employs the Stemming technique in combination with a string-matching method to handle ambiguous words and apply similarity levels using the Croft metric to improve the performance of the disambiguation level. After running a range of tests, the performance of the suggested approach was assessed with precision, recall, and F1-score metrics. It reached 78% of precision, 64% of recall, and 71% of F1-score, outperforming the traditional methods, such as the Lesk algorithm, which showed 65% of precision and 58% of recall. The further 5-fold cross-validation analysis was run to ensure the stability of the results. The average value of precision was 78% + -2%, which means the low range of the values in different validation sets. It allowed to state the ultimate performance stability of the proposed algorithm. The findings indicated that the best disambiguation level was achieved using a three-word context window, which provided a high level of balance between precision and the context’s size. The conclusion is that the proposed algorithm has a higher level of performance of the Arabic disambiguation than the existing knowledge-based approaches and can reach more than 70% of F1-score, showing the increased levels of precision. At the same time, one of the major drawbacks is the lower level of recall due to the high sensibility of the algorithms to the size of the context and the necessity to use pre-labeled datasets. One of the further research trends is a deeper analysis of the algorithm to choose the optimal context size and run the deep learning that can reach the improvement of the recall rate.