Modeling energy band gap of doped TiO2 semiconductor using homogeneously hybridized support vector regression with gravitational search algorithm hyper-parameter optimization

Titanium dioxide (TiO2) semiconductor is characterized with a wide band gap and attracts a significant attention for several applications that include solar cell carrier transportation and photo-catalysis. The tunable band gap of this semiconductor coupled with low cost, chemical stability and non-toxicity make it indispensable for these applications. Structural distortion always accompany TiO2 band gap tuning through doping and this present work utilizes the resulting structural lattice distortion to estimate band gap of doped TiO2 using support vector regression (SVR) coupled with novel gravitational search algorithm (GSA) for hyper-parameters optimization. In order to fully capture the non-linear relationship between lattice distortion and band gap, two SVR models were homogeneously hybridized and were subsequently optimized using GSA. GSA-HSVR (hybridized SVR) performs better than GSA-SVR model with performance improvement of 57.2% on the basis of root means square error reduction of the testing dataset. Effect of Co doping and Nitrogen-Iodine co-doping on band gap of TiO2 semiconductor was modeled and simulated. The obtained band gap estimates show excellent agreement with the values reported from the experiment. By implementing the models, band gap of doped TiO2 can be estimated with high level of precision and absorption ability of the semiconductor can be extended to visible region of the spectrum for improved properties and efficiency.