[{"data":1,"prerenderedAt":111},["ShallowReactive",2],{"glossary-en-overfitting":3},{"id":4,"title":5,"body":6,"description":100,"extension":101,"meta":102,"navigation":106,"path":107,"seo":108,"stem":109,"__hash__":110},"en_glossary\u002Fen\u002Fglossary\u002Foverfitting.md","What are Overfitting and Generalisation?",{"type":7,"value":8,"toc":89},"minimark",[9,14,22,27,33,37,59,63,74,78,81],[10,11,13],"h2",{"id":12},"what-is-overfitting","What is Overfitting?",[15,16,17,21],"p",{},[18,19,20],"strong",{},"Overfitting"," is when a model memorises details and noise specific to its training data instead of general patterns. The training score keeps rising while the validation score stalls or drops.",[23,24,26],"h3",{"id":25},"generalisation","Generalisation",[15,28,29,30,32],{},"The real goal is ",[18,31,25],{},": working well on data the model has never seen. In industrial projects that means a new field, a new machine or next month.",[23,34,36],{"id":35},"countermeasures","Countermeasures",[38,39,40,47,53],"ul",{},[41,42,43,46],"li",{},[18,44,45],{},"Early stopping",": stop training when the validation score has not improved for a set number of epochs.",[41,48,49,52],{},[18,50,51],{},"Data augmentation",": enrich the data with transforms such as rotation and brightness changes.",[41,54,55,58],{},[18,56,57],{},"More, and more diverse, data",": often the most effective fix.",[23,60,62],{"id":61},"data-leakage","Data Leakage",[15,64,65,66,69,70,73],{},"If information from training leaks into the test data, the score looks better than it really is. A common mistake in imaging projects is tiling a large orthophoto and splitting the tiles at random: neighbouring tiles are almost identical. The fix is to split the test by ",[18,67,68],{},"geographic blocks"," or by ",[18,71,72],{},"time",".",[23,75,77],{"id":76},"class-imbalance","Class Imbalance",[15,79,80],{},"If one class has very few examples (e.g. a handful of almond trees against hundreds of walnuts), that class scores low. This is a data problem, not a model fault; the fix is collecting more examples of that class.",[15,82,83,84],{},"Related: ",[85,86,88],"a",{"href":87},"\u002Fen\u002Fglossary\u002Fmodel-evaluation-metrics","Model Evaluation Metrics",{"title":90,"searchDepth":91,"depth":91,"links":92},"",2,[93],{"id":12,"depth":91,"text":13,"children":94},[95,97,98,99],{"id":25,"depth":96,"text":26},3,{"id":35,"depth":96,"text":36},{"id":61,"depth":96,"text":62},{"id":76,"depth":96,"text":77},"A model memorising its training data, data leakage, class imbalance and why measuring on unseen sites matters.","md",{"tags":103},[104,20,26,62,105],"Machine Learning","Industrial AI",true,"\u002Fen\u002Fglossary\u002Foverfitting",{"title":5,"description":100},"en\u002Fglossary\u002Foverfitting","zFWKlkN0Iz_cABqL7WUvwjt160D1zeI8ro00moUpavQ",1793875325092]