The performance of interacting metal surfaces having sliding motion is enhanced by fabricating microfeatures on-top of the surfaces. Laser powder bed fusion (LPBF) is used to fabricate the microfeatures of IN625 on-top of a flat surface. In LPBF, geometrical characteristics of microfeatures are based on the energy supplied and the volume of material melted (layer thickness) at that instant of time. There is very few research reported in open literature on the effect of layer thickness on microfeatures geometry during LPBF. In this work, a series of single tracks is fabricated with different combinations of laser power and scanning speed for 75 μm, 100 μm and 125 μm layer thickness and its effect on single track geometrical characteristics and microstructure was evaluated. It has been found that laser power and scanning speed have a significant influence on the track geometry irrespective of layer thickness. The minimum linear energy density of 2.5 J/mm is considered to fabricate the continuous track formation for investigated layer thickness. From statistical analysis, it is observed that the laser power and scanning speed shows higher contribution on track width and remelting depth, respectively. Whereas, layer thickness shows a higher influence on track height and contact angle of single track as compared to laser power and scanning speed. In 125 μm layer thickness, the volume of material melting is increased which encourages the gravity effect to deform the melt pool and produces larger track width compare to 100 μm layer thickness. Further, the cell spacing is gradually increased with layer thickness resulting increase in the solidification period. The shape controllability of the single track is significantly increased with layer thickness, but as keeps on increasing layer thickness due to the larger volume of molten metal, the melt pool losses ability to maintain convex shapes. An Artificial Neural Network (ANN) model is developed to understand the relationship between the LPBF process parameters and single track geometrical characteristics to predict the responses for a wide range of process parameters. The multiple response neural network model shows excellent agreement with experimental results with R2 value greater than 0.95. © 2024 Elsevier Ltd