Background: We aimed to develop novel diagnostic and prognostic signatures based on preoperative inflammatory, immunological, and nutritional parameters in blood (PIINPBs) by machine learning algorithms for patients with oral squamous cell carcinoma (OSCC). Methods: A total of 486 OSCC patients and 200 age and gender-matched non-OSCC patients who were diagnosed and treated at our institution for noninfectious, nontumor diseases were retrospectively enrolled and divided into training and validation cohorts. Based on PIINPB, 6 machine learning classifiers including random forest, support vector machine, extreme gradient boosting, naive Bayes, neural network, and logistic regression were used to derive diagnostic models, while least absolute shrinkage and selection operator (LASSO) analyses were employed to construct prognostic signatures. A novel prognostic nomogram integrating a PIINPB-derived prognostic signature and selected clinicopathological parameters was further developed. Performances of these signatures were assessed by receiver operating characteristic (ROC) curves, calibrating curves, and decision tree. Results: Diagnostic models developed by machine learning algorithms from 13 PIINPBs, which included counts of white blood cells (WBC), neutrophils (N), monocytes (M), lymphocytes (L), platelets (P), albumin (ALB), and hemoglobin (Hb), along with albumin-globulin ratio (A/G), neutrophil-lymphocyte ratio (NLR), platelet-lymphocyte ratio (PLR), lymphocyte-monocyte ratio (LMR), systemic immune-inflammation index (SII), and prognostic nutritional index (PNI), displayed satisfactory discriminating capabilities in patients with or without OSCC, and among OSCC patients with diverse pathological grades and clinical stages. A prognostic signature based on 6 survival-associated PIINPBs (L, P, PNI, LMR, SII, A/G) served as an independent factor to predict patient survival. Moreover, a novel nomogram integrating prognostic signature and tumor size, pathological grade, cervical node metastasis, and clinical stage significantly enhanced prognostic power [3-year area under the curve (AUC) =0.825; 5-year AUC =0.845]. Conclusions: Our results generated novel and robust diagnostic and prognostic signatures derived from PIINPBs by machine learning for OSCC. Performance of these signatures suggest the potential for PIINPBs to supplement current regimens and provide better patient stratification and prognostic prediction.