Optimal　foraging　algorithm　(OFA)　is　a　newly　stochastic　optimization　technique　and　is　famous　for　its　computational　accuracy.　However,　the　high　computational　accuracy　leads　to　slow　convergence　speed.　Experimental　results　demonstrate　that　OFA　is　good　at　unimodal　functions　but　poor　at　multimodal　functions.　To　improve　these　drawbacks,　in　this　paper　a　novel　modified　OFA　with　direction　prediction　and　Gaussian　oscillation,　named　OFA/P&G　is　introduced.　In　OFA/P&G,　a　transition　matrix　is　constructed　when　a　new　global　optimum　is　found　to　generate　the　candidate　individuals.　If　the　current　global　optimum　does　not　change,　the　Gaussian　oscillation　is　employed　in　a　low　probability　and　OFA　update　method　is　used　in　a　high　probability　to　generate　the　candidate　individuals.　The　superior　performance　of　OFA/P&G　is　verified　on　the　12　CEC2017　benchmark　functions,　13　constrained　benchmark　functions　and　5　engineering　problems.　Experimental　results　demonstrate　that　OFA/P&G　outperforms　other　comparative　algorithms.　Finally,　a　real-world　problem,　drilling　path　optimization,　is　solved　by　OFA/P&G.　©　2022　Elsevier　Ltd