Machine translation (MT) is a computer science field that facilitates text translation between languages in sectors like education, tourism, and trade with the advantages of speed and cost-effectiveness. However, its drawback is low accuracy due to language structure differences. Currently, two MT approaches exist: classical and neural machine translation (NMT). NMT, employing deep recurrent neural networks (RNNs), aims for higher accuracy. The African continent has the lowest utilization rate of MT in the world. Existing African language MTs are based on the classical approach. The primary technical reason for the low MT adoption is the lack of accurate African languages MT. This research developed an NMT model using stacked Long Short-Term Memory (LSTM) architecture to translate English to Kichuka. The dataset (KED21) comprised 5000 Kichuka to English sentences and words from Chuka University Humanities, split into 70% training, 20% testing, and 10% validation. Evaluation compared the LSTM and statistical-based MT (SBMT) models using BLEU scores. LSTM achieved 96% translation accuracy, outperforming SBMT at 14%, with better precision, recall, F1, and mean average weighted precision (96%, 96%, 95%, and 93% respectively). External validity tested LSTM with the Luganda to English (LED21) dataset, achieving 92% accuracy, compared to SBMT 13%, with improved precision, recall, F1, and mean average weighted precision (92%, 95%, 94%, and 92%). Results indicate that the LSTM NMT model significantly improves translation accuracy, showing promise for enhancing MT adoption in African languages.