1. The Problem Statement

2. Materials and Methods

2.1. Data Collection and Pre-processing

The data utilized in this study pertains to the automotive service parts sector in Thailand. The dataset encompasses detailed information on part counts, new model parts, sales records, lead times, and vehicle volumes. Specifically, the data encompasses nine months of sales data from January 2022 to September 2022, and forecasts for October 2022 to February 2023. For the purposes of this research, the domestic market sales data, excluding export transactions, were analysed, as these were deemed most pertinent.

Table 1. Description of the dataset used in the study, including the total counts for different commodity categories.

Table 1. Description of the dataset used in the study, including the total counts for different commodity categories.

Commodity	Total
Accessories	872
Collision	20426
Light Repair	3623
Maintenance	5643
Other	353
Powertrain	1912
Grand Total	32829

The initial dataset was subjected to a comprehensive cleansing process to ensure its suitability for analysis. Records reflecting negative stock levels were excluded, and any missing values were replaced with zeros. In addition, parts that could be categorized as Periodic Maintenance Service Parts and Accessories were excluded from the dataset to focus solely on core service parts.

2.2. Data Categorization

In alignment with the newly established demand criteria rules, the parts were categorized. Only parts falling within the top 25%, referred to as “Platinum,” “Diamond,” “Gold,” “Silver,” and “Bronze” categories, were considered for the subsequent analysis. The criteria for categorization were based on demand and order frequency.

In the Table 2, we observe the number of service parts falling within each demand category and their corresponding percentages. It becomes evident that the majority of service parts are distributed across the “Silver” and “Bronze” categories, with 29.3% and 44.7% representation, respectively. Notably, the “Platinum” category contains a comparatively smaller share at 16.2%, while the “Diamond” and “Gold” categories are at 9.1% and 9.7%, respectively. The total number of service parts considered for this analysis is 1806.

Table 2. Categorization of service parts based on demand criteria, including counts and percentages.

Table 2. Categorization of service parts based on demand criteria, including counts and percentages.

Category	Count	%
Platinum	128	16.2%
Diamond	165	9.1%
Gold	176	9.7%
Silver	530	29.3%
Bronze	807	44.7%

This distribution has significant implications for strategic decision-making. The differentiation between these categories allows for tailored resource allocation and focus. High-demand service parts categorized as “Platinum” and “Diamond” are prime candidates for heightened inventory management efforts, ensuring their availability to minimize service disruptions. On the other hand, the “Bronze” category, comprising low-demand parts, offers opportunities for alternative, cost-effective solutions.

2.3. Decision Rules for Stocking

In the intricate landscape of service parts management, the decision on which parts to stock involves a judicious evaluation of various factors. The criteria for stocking service parts are not arbitrary; instead, they are meticulously derived from key parameters such as purchase cost (CST), annual demand (Q), and backorder cost (B). These factors play a crucial role in the development of a structured and efficient decision-making framework [7].

2.3.1. Decision Rules Formulation

The decision to stock service parts is contingent on a set of indices, each carefully formulated to capture specific aspects of the stocking decision:

• Cost Index (CSTi): This index (CSTi) is calculated using the formula CSTi = - log2(CST/3000). It quantifies the cost efficiency of stocking a particular service part, considering the purchase cost.
• Demand Index (Qi): The demand index (Qi) is formulated as Qi = - log2(Q/10), where Q represents the annual demand for a service part. This index provides insights into the demand dynamics, emphasizing the significance of considering the annual demand in the decision-making process.
• Backorder Index (Bi): The backorder index (Bi) is determined by the formula Bi = log2(B/18000), with B representing the backorder cost. This index focuses on the cost implications associated with potential backorders, contributing a crucial dimension to the decision rules.

The decision rule is framed as: Stock the part if

CSTi + Qi + Bi > 0

To provide a comprehensive overview of the decision rules, Table 3 outlines the decision rules applied to determine whether specific service parts should be stocked or not. The formulation of these decision rules involved a meticulous consideration of various factors, including Lead Time (LT), Purchase Cost (CST), Classification Criteria, and Backorder Cost. The objective was to establish a systematic approach that optimally balances the costs associated with stocking and the potential backorder expenses.

Table 3 outlines a few samples where the Sum Index is < 0. Parts with high cost, low to medium demand criteria combined with lower lead time fall under the Do not stock category. The table presents specific decision rules for individual service parts, including information such as Service Part Number (SPN), Commodity, Lead Time (LT), Purchase Cost (CST), criteria, Backorder Cost, and the final decision (Stock or Do Not Stock). The columns provide detailed information on key parameters:

• Index: An index assigned to each service part for reference in the decision-making process.
• Decision: Indicates whether the service part should be stocked or not based on the decision rules.
• LT (Lead Time): The time it takes for the delivery of the service part after placing an order.
• CST (Purchase Cost): The cost associated with acquiring the service part.
• Criteria: The classification criteria used in the decision-making process, such as “Critical,” “low-high,” or “medium-high.”
• Backorder Cost: The cost incurred if the service part is not stocked and needs to be backordered.
• Sum: A cumulative value derived from specific calculations, providing an overall assessment of the decision-making criteria.

Table 3. Decision Rules for Stocking Service Parts.

Table 3. Decision Rules for Stocking Service Parts.

SPN	Commodity	LT	CST	Criteria	Backorder Cost	Sum Index	Decision
SPN_00105	LIGHT REPAIR	78	409.829	Critical	93636.88	6	Stock
SPN_00476	LIGHT REPAIR	135	371,513	medium-high	162033.44	6	Stock
SPN_00502	LIGHT REPAIR	66	110.223	medium-high	79209.92	6	Stock
SPN_00562	LIGHT REPAIR	180	445,659	medium-high	216040.11	6	Stock
SPN_00606	LIGHT REPAIR	15	7191.13	medium-high	18071.91	-1	Do Not Stock
SPN 00634	LIGHT REPAIR	90	227.22	medium-high	108020.45	6	Stock
SPN_00719	LIGHT REPAIR	66	10470.75	low-high	80142.37	-1	Do Not Stock

The decision rules were derived through a combination of historical data analysis and strategic considerations to ensure the efficient allocation of resources and minimize costs. The “Decision” column reflects the outcome of this comprehensive evaluation, guiding the stocking strategy for each service part. The “Sum” column serves as a consolidated indicator, aggregating relevant criteria for a holistic decision-making perspective. This table serves as a valuable reference for understanding the nuanced decision-making process involved in optimizing service parts management within the automotive industry.

2.3.2. Visualization

To enhance the clarity of the decision-making process, a visual representation in the form of a diagram or flowchart has been incorporated in Figure 2. This visualization serves as a guide, illustrating the sequential steps involved in determining whether a service part should be stocked or not. It encapsulates the logic and criteria embedded in the decision rules, providing a comprehensive overview of the stocking process. The flowchart (Figure 2) visually represents the decision-making process for stocking service parts. It likely outlines the steps involved in evaluating different criteria and making the final decision on whether to stock a particular service part or not.

This visualization not only aids in understanding the intricate decision rules but also serves as a valuable reference for stakeholders involved in service parts management. It streamlines the decision-making process, making it more transparent and accessible for effective implementation [7]. The subsequent sections will delve into the specifics of these decision rules, providing a deeper insight into how each criterion contributes to the overall stocking strategy.

2.4. Forecasting Models

In the pursuit of identifying an effective forecasting model for service parts demand, a diverse set of approaches was explored. This section outlines the application of various models, ranging from classical time series methods to advanced machine learning techniques.

2.4.1. Model Selection and Implementation

The selection of forecasting models involved a comprehensive consideration of different approaches. Classical time series models, including Moving Average (MA), Weighted Moving Average (WMA), and Exponential Smoothing, were among the initial methodologies applied [9]. These traditional models provided a baseline for comparison against more sophisticated techniques [10].

In addition to classical methods, machine learning models were introduced to harness the power of data-driven insights. Support Vector Regressor (SVR) [11] with Long Short-Term Memory (LSTM) [12] networks, known for their ability to capture complex patterns, were employed for demand forecasting. To ensure a comprehensive evaluation, Prophet, Auto ARIMA and Histogram-based Gradient Boosting Regressor models were also incorporated.

2.4.2. Performance Evaluation

Quantifying the performance of each forecasting model is critical for selecting the most reliable approach. The evaluation criteria primarily revolved around the Mean Absolute Percentage Error (MAPE) [13]. This metric provides a measure of the accuracy of each model by assessing the percentage difference between predicted and actual values.

Table 4 summarizes the preliminary results obtained from the application of different forecasting models, presenting their respective Mean Absolute Percentage Error (MAPE). This table serves as a quick reference to compare the performance of each model, with lower MAPE values indicating higher accuracy.

Table 4. Preliminary results from applying different forecasting models.

Table 4. Preliminary results from applying different forecasting models.

Model	Mean Absolute Percentage Error (MAPE)
Moving Average	44.6%
Exponential Smoothing	36.3%
Weighted Moving Average	40.7%
Support Vector Machine (SVM)	18.1%
Prophet	27.6%
Auto ARIMA	29.9%
Histogram-based Gradient Boosting	29.1%

The subsequent analysis and decision-making will be guided by these initial insights, aiming to identify the most effective forecasting model for service parts demand.

3. Results and Discussion

3.1. Service Parts Classification Based on Demand Criteria

3.1.1. Demand-Based Categories

In line with the proposed criteria, the service parts were successfully categorized into five distinct groups: “Platinum,” “Diamond,” “Gold,” “Silver,” and “Bronze.” The Distribution of parts across these categories is presented in Table 5. This classification provides a structured foundation for making informed decisions about stocking service parts.

Table 5. Distribution of Service Parts Based on Demand Categories.

Table 5. Distribution of Service Parts Based on Demand Categories.

Demand Category	Number of Parts	Percentage (%)
Platinum	128	16.2%
Diamond	165	9.1%
Gold	176	9.7%
Silver	530	29.3%
Bronze	807	44.7%
Total	1806	100.0%

Figure 3 displays a histogram plot that provides valuable insights into the distribution of service parts based on certain criteria. The x-axis represents the criteria, while the y-axis represents the number of service parts. Notably, this plot illustrates a significantly higher number of parts with low demand. Approximately 75% of the service parts recorded sales of fewer than five pieces in the initial eight months post-launch. This observation is particularly crucial for understanding the demand life cycle of service parts, as it highlights the challenges posed by the overestimation of demand in the initial phase. The distribution of demand across various categories becomes apparent in this representation, which is instrumental in devising optimal stocking strategies.

Figure 4 presents a violin plot that focuses on the distribution of service parts across various part systems. The x-axis distinguishes different part systems, while the y-axis indicates the density of parts in each system. Notably, this plot reveals the presence of outliers in three distinct systems, namely Instrument Panel & Console, Powertrain, and Underbody Metal. While Powertrain might suggest the inclusion of Periodic Maintenance Service parts, the implications of outliers in the Instrument Panel & Console and Underbody Metal categories warrant further investigation. The identification of these outliers within part systems is a vital step in optimizing stocking decisions and resource allocation, ensuring that the right parts are stocked efficiently.

3.1.2. Implications of Categorization

The categorization of service parts based on demand criteria has several notable implications [14]. Notably, it allows for a more strategic allocation of resources. The “Platinum” and “Diamond” categories, representing high-demand parts, warrant a more significant focus on inventory management. In contrast, the “Bronze” category, with low-demand parts, opens opportunities for alternative solutions. The strategic implications of this classification are discussed in Section 3.2.

3.2. Stocking Decision Rules and Outcomes

3.2.1. Decision Rule Application

Applying the decision rules established in Section 3.1, the study determined whether to stock each service part based on the calculated indices. The outcomes are presented in Table 6. A significant number of parts were identified as candidates for stocking, aligning with the management’s objective to stock at least one piece of every part [15]. Notably, this aligns with the aim of minimizing service disruptions. The outcomes are summarized in Table 3.1.

Table 6. Decision Rule Outcomes.

Table 6. Decision Rule Outcomes.

Classification	Do Not Stock	Stock
low-low	294	454
low-high	74	221
medium-high	6	214
high-high	3	160
low-medium	2	57
high-low		13
medium-low		15
Critical		128
very-high		165
Grand Total	379	1427

3.2.2. Exceptions and Contingency Solutions

Intriguingly, a set of exceptions and contingency solutions was introduced. Based on factors such as classification criteria and lead times, exceptions were recommended for certain parts that initially fell into the stocking category. Similarly, contingency solutions focused on minimizing costs by addressing parts with high lead times and low demand. These strategies introduce a level of flexibility into the decision-making process, ensuring the optimal allocation of resources.

4. Conclusions

References

1. Kato, A. I., & Manchidi, N. M. (2022). Impact of supply chain management strategies on firms’ sustainable performance: a case of an emerging economy. Entrepreneurship and Sustainability Issues, 10(2), 93. [CrossRef]
2. Jiao, R., Commuri, S., Panchal, J., Milisavljevic-Syed, J., Allen, J. K., Mistree, F., & Schaefer, D. (2021). Design engineering in the age of industry 4.0. Journal of Mechanical Design, 143(7), 070801. [CrossRef]
3. Cardeal, G., Leite, M., & Ribeiro, I. (2023). Decision-support model to select spare parts suitable for additive manufacturing. Computers in Industry, 144, 103798. [CrossRef]
4. Ding, S., & Li, R. (2021). Forecasting the sales and stock of electric vehicles using a novel self-adaptive optimized grey model. Engineering Applications of Artificial Intelligence, 100, 104148. [CrossRef]
5. Aunyawong, W., Wararatchai, P., & Hotrawaisaya, C. (2020). The influence of supply chain integration on supply chain performance of auto-parts manufacturers in Thailand: a mediation approach. International Journal of Supply Chain Management, 9(3), 578-590.
6. Min, S., Zacharia, Z. G., & Smith, C. D. (2019). Defining supply chain management: in the past, present, and future. Journal of business logistics, 40(1), 44-55. [CrossRef]
7. Yang, Ke, Yongjian Wang, Shidong Fan, and Ali Mosleh. 2021. "Multi-Criteria Spare Parts Classification Using the Deep Convolutional Neural Network Method" Applied Sciences 11, no. 15: 7088. [CrossRef]
8. Jian, J., Li, B., Zhang, N., & Su, J. (2021). Decision-making and coordination of green closed-loop supply chain with fairness concern. Journal of Cleaner Production, 298, 126779. [CrossRef]
9. Alex, M. A. H., & Rahmawati, N. (2023). Application of the Single Moving Average, Weighted Moving Average and Exponential Smoothing Methods For Forecasting Demand At Boy Delivery. Tibuana, 6(1), 32-37. [CrossRef]
10. ArunKumar, K. E., Kalaga, D. V., Kumar, C. M. S., Chilkoor, G., Kawaji, M., & Brenza, T. M. (2021). Forecasting the dynamics of cumulative COVID-19 cases (confirmed, recovered and deaths) for top-16 countries using statistical machine learning models: Auto-Regressive Integrated Moving Average (ARIMA) and Seasonal Auto-Regressive Integrated Moving Average (SARIMA). Applied soft computing, 103, 107161. [CrossRef]
11. Huang, J., Algahtani, M., & Kaewunruen, S. (2022). Energy forecasting in a public building: a benchmarking analysis on long short-term memory (LSTM), support vector regression (SVR), and extreme gradient boosting (XGBoost) networks. Applied Sciences, 12(19), 9788. [CrossRef]
12. Weng, T., Liu, W., & Xiao, J. (2020). Supply chain sales forecasting based on lightGBM and LSTM combination model. Industrial Management & Data Systems, 120(2), 265-279. [CrossRef]
13. Kilimci, Z. H., Akyuz, A. O., Uysal, M., Akyokus, S., Uysal, M. O., Atak Bulbul, B., & Ekmis, M. A. (2019). An improved demand forecasting model using deep learning approach and proposed decision integration strategy for supply chain. Complexity, 2019. [CrossRef]
14. Chandriah, K. K., & Naraganahalli, R. V. (2021). RNN/LSTM with modified Adam optimizer in deep learning approach for automobile spare parts demand forecasting. Multimedia Tools and Applications, 80(17), 26145-26159. [CrossRef]
15. Agrawal, N., & Deshpande, V. (2022). Service Parts Management: Theoretical Foundations, Practice, and Opportunities. Creating Values with Operations and Analytics: A Tribute to the Contributions of Professor Morris Cohen, 133-167. [CrossRef]
16. Muniz, L. R., Conceição, S. V., Rodrigues, L. F., de Freitas Almeida, J. F., & Affonso, T. B. (2021). Spare parts inventory management: a new hybrid approach. The International Journal of Logistics Management, 32(1), 40-67. [CrossRef]
17. Shahriar, S. A., Kayes, I., Hasan, K., Hasan, M., Islam, R., Awang, N. R., ... & Salam, M. A. (2021). Potential of ARIMA-ANN, ARIMA-SVM, DT and CatBoost for atmospheric PM2. 5 forecasting in Bangladesh. Atmosphere, 12(1), 100. [CrossRef]
18. Amirvaresi, A., & Parastar, H. (2021). External parameter orthogonalization-support vector machine for processing of attenuated total reflectance-mid-infrared spectra: A solution for saffron authenticity problem. Analytica Chimica Acta, 1154, 338308. [CrossRef]