Predicting Small Business Longevity and Competitiveness Through Self-Random Manhattan Botox-Coupled Attention Network Driven Strategic Entrepreneurship

Abstract

Predicting the survival and competitiveness of small businesses is an intricate problem owing to the multidimensional nature of strategic, financial, and operational aspects. Most current models do not accurately capture the complex dynamics between firm dynamics and survival patterns over the long run. In view of this, the study presents a new deep learning model known as Self random Manhattan Botox coupled Attention network (Self-RanMB-CAN), whose goal is to improve the accuracy of prediction for small business survival and competitiveness. A structured dataset named "Small Business Longevity Dataset" consisting of 500 records with 12 critical features has been created to enable this analysis. The data is preprocessed with Cumulative Curve Fitting Approximation (CCFA) to provide normalized trend representation, and then the Discrete Cosine-Krawtchouk–Tchebichef Transform (DCKTKT) is applied for feature extraction to maintain spectral and spatial information. The Self-RanMB-CAN model proposed incorporates a Random-Coupled Neural Network (RCNN) coupled with Manhattan Self- Attention (MaSA), enabling targeted attention on high-impact features, while the Botox Optimization Algorithm (BOA) optimally adapts network parameters to drive performance to its maximum potential. Experimental results reflect a prediction accuracy level of 99.9%, reflecting the stability of the proposed approach. The proposed model has twofold advantages: (i) it greatly improves feature discrimination for better strategic decision-making and (ii) guarantees optimal learning of parameters in complex business environments.