A CNN-LSTM Based Predictive Model for Intelligent Efficiency Assessment in Industry Environments

Precise prediction of production efficiency is necessary to maximize production quality, reduce downtime, and facilitate efficient smart factory operations in Industry 4.0 settings. Current prediction models are challenged by industrial multi-modal data, high feature variability, and non-transparency in decision-making. This work proposes EffiNet, a tailored deep neural model for multi-class efficiency status prediction—ranking the manufacturing performance to Low, Medium, and High levels of efficiency. Taking advantage of the Intelligent Manufacturing Dataset for Predictive Optimization which combines Industrial IoT sensor measurements (temperature, vibration, power consumption) with 6G network metrics (latency, packet loss, communication rate), EffiNet obtains high-fidelity representations of process behavior with the help of hierarchical dense layers with batch normalization, adaptive dropout, and L1–L2 regularization in order to ensure stability. The model utilizes balanced resampling and robust scaling in order to address class imbalance and ensure stable learning dynamics. Empirical performance indicates excellent predictive capability with overall accuracy of 93.08% and weighted F1-score of 0.93 with per-class accuracies of 95.9%, 86.3%, and 97.0% for low, medium, and high efficiency classes, respectively. The results indicate the capability of EffiNet to model nonlinear relationships between production parameters and network metrics for AI-based predictive optimization of 6G-capable manufacturing systems. The research promotes the development of smart, data-oriented, and transparent efficiency measurement systems in future industrial operations.