This paper presents an evaluation system for the effectiveness of forehand table tennis strokes using a bidirectional long short-term memory (BLSTM) model optimized with an improved dragonfly algorithm (IDFOA). The system aims to provide a virtual coach that can assess and provide feedback on players' performance, particularly during home-based sports activities during the COVID-19 lockdown. The study collects data from 16 players using a miniaturized inertial sensor (BNO055) attached to a table tennis racket, tracking their forehand strokes and assigning ratings based on three instructors' input. The BLSTM-IDFOA model is trained to predict stroke quality scores based on sensory data, with hyperparameters optimized using the IDFOA. The results show that the BLSTM-IDFOA model outperforms other regression methods, achieving lower RMSE, MAE, MAPE, and higher R2 values compared to LSTM and other models. The study concludes that the proposed method is effective in assessing table tennis stroke quality and can be integrated with cloud computing services for continuous player improvement. Future work includes enhancing the system with cloud services and visual feedback for coaches.This paper presents an evaluation system for the effectiveness of forehand table tennis strokes using a bidirectional long short-term memory (BLSTM) model optimized with an improved dragonfly algorithm (IDFOA). The system aims to provide a virtual coach that can assess and provide feedback on players' performance, particularly during home-based sports activities during the COVID-19 lockdown. The study collects data from 16 players using a miniaturized inertial sensor (BNO055) attached to a table tennis racket, tracking their forehand strokes and assigning ratings based on three instructors' input. The BLSTM-IDFOA model is trained to predict stroke quality scores based on sensory data, with hyperparameters optimized using the IDFOA. The results show that the BLSTM-IDFOA model outperforms other regression methods, achieving lower RMSE, MAE, MAPE, and higher R2 values compared to LSTM and other models. The study concludes that the proposed method is effective in assessing table tennis stroke quality and can be integrated with cloud computing services for continuous player improvement. Future work includes enhancing the system with cloud services and visual feedback for coaches.