Artificial intelligent-enhanced federated approach for employee performance evaluation using multimodal data

Evaluating employee performance is important for getting the most out of any workforce, traditional appraisal methods often depend on subjective judgments and a small number of data sources. Research shows to use a neural network to evaluate employee performance by combining behavioral logs, textual feedback, and speech or audio signals into a single system. Multi-Modal system puts together different types of data, like quantitative performance measures, written feedback, behavioral logs, and speech or audio signals, into one evaluation framework. Spectral noise reduction improves the audio data by reducing interference. Textual and behavioral data are cleaned using stop-word removal and normalization to reduce redundancy and improve clarity. Feature extraction is performed using Mel-frequency cepstral coefficients for audio characteristics, and Word2Vec embeddings for semantic understanding of textual feedback. Principal component analysis is employed to extract and compress high-dimensional numerical behavioral features. The extracted features are fused and then fed into the evaluation system. The green anaconda optimized federated neural network with attention model evaluates employee performance, and it optimizes the neural network's weights and parameters to improve convergence and overall prediction performance and hyperparameter tuning. Results prove that the multimodal model significantly outperforms single-modality approaches, achieving 98.82% accuracy, enabling actionable insights for managers. The system was implemented using Python 3.11 for model training, evaluation, and performance visualization. It highlights the potential of neural network–driven multimodal data integration to bridge the gap between subjective evaluations and objective performance measurement, fostering transparency, fairness, and productivity in employee appraisal processes.