The primary goal of this paper is to enhance the capacity while minimizing visual distortion and improving secrecy. To accomplish this task, a novel information concealing approach based on Machine Learning (ML) block classifier is proposed to improve performance. ML techniques are used to classify the blocks of the image using extracted features. The concealing capacity of each block is decided depending on the type of each block. Secret messages are compressed using the Adaptive Huffman encoding method prior to embedding. The proposed model is validated in three cases: performance of the ML classifier, efficiency of embedding method and its robustness. Deep Neural Network (DNN) block classifier exhibits 99% validation accuracy which is significantly better than the logistic regression and random forest classifier. Embedding method using DNN as the block classifier achieved PSNR more than 50 with embedding rate 0.5 bpp. The proposed scheme is robust and sustain