Mechanical properties of fine grain nanocomposites differ from those of conventional composites due to the in situ effect caused by the addition of nanoparticle reinforcement and the complexity of strengthening mechanisms, which make their prediction using conventional analytical and numerical model is relatively difficult. Therefore, this work presents a rapid reliable machine learning model based on long-short term memory model modified with beluga whale optimizer to predict the mechanical properties of ultrafine grain Al-TiO2 nanocomposite manufactured using accumulative roll bonding (ARB). The mechanical properties were evaluated using tensile tests and correlated with the composite microstructure and hardness. Experimentally, it was demonstrated that the tensile strength increases with increasing the number of ARB passes until a plateau was achieved due to the uniform distribution of TiO2 nanoparticles inside the composite and the saturation of grain refinement in the Al matrix. The maximum tensile achieved was 270 MPa for composite containing 3% TiO2 nanoparticles after 5 ARB passes compared to 90.5 MPa for the raw Al. The proposed model was able to predict the yield and ultimate strengths, elongation, and hardness for all the produced composites tested with excellent accuracy reaching R2 equal 0.9955, 0.9952, 0.9859, and 0.9975, respectively, which is way better than other models.