← A Novel Machine Learning Approach for Data Analysis and Prediction

← Back To Current Issues

A Novel Machine Learning Approach for Data Analysis and Prediction

Elena V. Marchetti1,2,#, Rajiv S. Bhatt1,#, Thomas O. Nkemelu1,2, Yuki Sato1, Claire M. Dubois1,2,*

Affiliations

1Department of Data Science and Predictive Analytics, Northfield Institute of Technology, Hallberg, Sweden; 2Centre for Applied Machine Learning Research, Northfield Institute of Technology, Hallberg, Sweden;

Correspondence

Dr. Claire M. Dubois, Department of Data Science and Predictive Analytics, Northfield Institute of Technology, Hallberg, Sweden; Email: c.dubois@northfieldit.example.edu

Abstract

Accurate prediction from heterogeneous, high-dimensional data remains a central challenge in applied machine learning, particularly when datasets combine structured tabular records with time-dependent signals. This study proposes a hybrid convolutional-recurrent architecture, combining a convolutional feature extractor, a long short-term memory (LSTM) temporal branch, and an attention mechanism, for joint analysis and prediction across mixed data modalities. The proposed model was evaluated against three established baselines (logistic regression, random forest, and gradient boosting) on a combined structured and time-series benchmark of 48,600 records. The hybrid model achieved 94.7% accuracy and an F1-score of 0.938, outperforming the strongest baseline (gradient boosting) by 5.5 percentage points, at a modest increase in inference time (14.3 ms versus 9.8 ms). Ablation analysis showed that removing the attention layer reduced accuracy by 3.4 points and removing the LSTM branch reduced accuracy by 6.1 points, indicating that both temporal modeling and attention-based feature weighting contribute meaningfully to overall performance. These results suggest that hybrid architectures combining convolutional, recurrent, and attention components offer a practical and accurate approach for prediction tasks involving mixed structured and sequential data.

Keywords

Machine learning · Data analysis · Predictive modeling · Deep learning · Attention mechanism · Hybrid neural architecture