This study presents an integrated approach combining experimental investigation, machine learning (ML), and response surface methodology (RSM) to assess and optimize the performance and emissions of a diesel engine fueled with low-percentage waste tire pyrolysis oil (TO) blends. Diesel–TO blends at 2 % (TO2) and 7 % (TO7) were tested alongside pure diesel (D100) across engine speeds from 1100 to 2400 rpm. Key metrics such as brake thermal efficiency (BTE), brake specific fuel consumption (BSFC), and emissions (NO<inf>x</inf>, CO<inf>2</inf>, HC, exhaust gas temperature) were measured. Three ML models Artificial Neural Network (ANN), Random Forest (RF), and eXtreme Gradient Boosting (XGBoost) were trained on 45 experimental data points to predict engine behavior. RSM was applied using a 13point design to model nonlinear interactions and perform multi-objective optimization. Results showed a maximum BTE of 33 % for D100 and 30 % for both TO2 and TO7. BSFC was lowest at 198 gkWh^-1 for D100, with slightly higher values for TO blends. TO7 exhibited peak NO<inf>x</inf> emissions of 640 ppm but showed HC reduction to 8 ppm at higher speeds. CO<inf>2</inf> emissions declined with speed, reaching 11.7 % for TO7 at 1800 rpm. Among ML models, RF and XGBoost achieved the best predictive accuracy, with most predictions within ±10 % of experimental values. RSM optimization identified 1.2 % TO at 2344 rpm as optimal, predicting BTE of 25.81 %, BSFC of 328.10 g/kWh, and NO<inf>x</inf> of 306.97 ppm. This study confirms that low-percentage TO blends offer viable engine performance with lower emissions, and that ML–RSM integration enhances predictive and optimization capabilities.