<p>Recently, Artificial intelligence (AI) techniques become popular in complicated healthcare applications, which aim to observe the large quantity of healthcare data to improve the quality of healthcare services. The AI models are commonly employed to inspect the huge volume of healthcare data, also for earlier disease detection and treatment. At the same time, malaria is a severe health issue caused by malaria parasites. The identification of malaria at the initial stage is necessary for efficient medications. Since the conventional detection process is time-consuming and expensive, the design of automated tools becomes essential to improve detection performance. To design an effective decision support system for the smart healthcare system, this paper presented a sparse autoencoder with an optimal deep neural network (SAE-ODNN) technique for the classification and detection of malaria parasites. The SAE-ODNN technique intends to determine the existence of malaria through blood smear images. Initially, pre-processing is performed by employing median filtering (MF) technique. To detect the infected regions, the U-Net-based segmentation method is utilized to segment the blood smear images. Besides, the SAE-ODNN approach utilizes SAE with smoothed regularization approach to generate a set of feature vectors. Moreover, the SAE-ODNN approach employs deep neural network (DNN) for classification. Furthermore, coyote optimization algorithm (COA) method is used to appropriately tune the hyperparameters of the DNN technique. A comprehensive series of simulation analyses was conducted on the SAE-ODNN method using the benchmark dataset. The obtained result demonstrates the enhanced performance of the SAE-ODNN method with a maximum detection accuracy of 98.82%.</p>