Accurate demand forecasting remains essential for optimizing textile supply chains and inventory management, yet model selection is challenging due to market volatility and data limitations. This study compares traditional, time series and machine learning approaches namely Seasonal Linear Trend, ARIMA, SARIMAX, XGBoost, Random Forest with respect to years of quarterly demand data of high moving textile raw materials. The historical sales data is secondary data collected from a leading Bangladeshi textile organization and sales forecasting is done for twelve such high moving and high consumable raw materials. Results are fascinating as for short-term operational sales forecasting in Bangladesh textile industry, Seasonal Linear outperforms all others with practical accuracy as for forecasting purposes, all approaches perform relatively close and rarely deviate but long-term sales forecasting for Random Forest and XGBoost approaches provide great potential for when historical data over the years will be available as this ensures stabilized maximum stock levels which would increase turnover by reducing overstock incidents and stockouts. Ultimately this study addresses some vital problems of Bangladeshi wholesale textiles; seasonal approaches sustain accuracy advantage in the short run for purchasing and production decisions while machine learning approaches possess similar advantages in the long run. This study contributes to the textile industry domain of demand forecasting by offering a comparison with substantive recommendations for improvement in accuracy and efficiency since it adds value to real life.