Description:FIELD OF INVENTION

[001] The present invention relates to inventory management, particularly, the present invention relates to unhealthy inventory management using self-learning machine learning models that determine existing surplus products and also predict the same for accurate demand forecasting.
BACKGROUND OF THE INVENTION
[002] Inventory management refers to the process of storing products in the warehouse, receiving orders from customers, and processing orders by delivering available products in the inventory. The goal of inventory management is to efficiently streamline inventories to avoid excess stock and shortages. Different methods are available for inventory management to determine any excess or shortage of stocks, such as manual analysis of inventory, ABC analysis, Sales trend analysis, Demand forecasting, Stockout prediction, Expiry date prediction, Lead time analysis, SKU rationalization, Seasonal demand planning, Inventory turn analysis, Price optimization, Slow moving inventory identification, Seasonal Adjusted Sales Analysis, Customer Demand Analysis, Stock Replenishment Forecasting, Product Life Cycle Analysis, Seasonal Adjusted Demand Forecasting, etc.
[003] In manual analysis, demand forecasting is based on available stocks and orders in hand, generally, business managers make a judgment on whether a product will sell well or not and take a call on buying inventory of products. However, manual judgment in determining sale prediction can be subjective and influenced by the personal biases of the business managers. This can lead to inaccurate predictions and mismanagement of inventory.
[004] Other inventory management methods such as ABC analysis categorize inventory into three groups based on their value and level of importance. For example, Group A items are high-value items that represent a small percentage of the total inventory but contribute a large percentage of the revenue. Group C items are low-value items that represent a large percentage of the total inventory but contribute a small percentage of the revenue. By analyzing the movement of inventory across these groups, businesses can identify items that are at risk of becoming unhealthy. Further, sales trend analysis involves analyzing historical sales data to identify trends and patterns in sales. By identifying products that are no longer selling well or are showing a decline in sales, businesses can identify which products are at risk of becoming unhealthy inventory. Whereas demand forecasting involves predicting future demand for products based on historical data and current trends. By accurately forecasting demand, businesses can avoid overstocking or understocking, which can lead to unhealthy inventory.
[005] However, said techniques provide a limited scope in determining sales trend analysis and demand forecasting which focus on specific aspects of inventory management. While they can provide valuable insights, they may not capture all the factors that contribute to unhealthy inventory, such as supply chain disruptions, changing market conditions, or customer behavior.
[006] Other prediction techniques such as stockout prediction involve using machine learning algorithms to predict the likelihood of a product running out of stock. By identifying potential stockouts, businesses can avoid overstocking and prevent unhealthy inventory. Moreover, expiry date prediction involves using data analytics to predict when products are likely to expire. By identifying products that are at risk of expiring soon, businesses can take proactive measures to sell or dispose of the products before they become unhealthy inventory. Whereas, lead time analysis involves analyzing the time it takes for a product to move from the supplier to the customer. By analyzing lead times, businesses can identify potential bottlenecks in the supply chain that may result in excess inventory or stockouts.
[007] Some additional inventory management techniques are:
[008] SKU rationalization: SKU rationalization involves analyzing the performance of individual stock-keeping units (SKUs) to identify which SKUs are performing well and which ones are not. By eliminating low-performing SKUs or combining them with other SKUs, businesses can reduce the risk of unhealthy inventory.
[009] Seasonal demand planning: Seasonal demand planning involves forecasting demand for products during specific seasons or events, such as holidays or sporting events. By accurately forecasting demand, businesses can avoid overstocking or understocking and reduce the risk of unhealthy inventory.
[0010] Inventory turn analysis: Inventory turn analysis involves analyzing how quickly inventory is sold and replaced. By analyzing inventory turns, businesses can identify which products are selling quickly and which ones are not. Products with low inventory turns may be at risk of becoming unhealthy inventory.
[0011] Price optimization: Price optimization involves using data analytics to determine the optimal price for products based on historical sales data and current market trends. By pricing products correctly, businesses can avoid overstocking or understocking and reduce the risk of unhealthy inventory.
[0012] Slow-moving inventory identification: Slow-moving inventory identification involves using data analytics to identify products that are not selling well or are taking longer to sell than expected. By identifying slow-moving inventory, businesses can take proactive measures to sell or dispose of the products before they become unhealthy inventory.
[0013] Seasonal Adjusted Sales Analysis: Seasonal Adjusted Sales Analysis is used to identify slow-moving inventory by adjusting sales data for seasonal factors. By identifying slow-moving inventory that may be seasonal, businesses can plan to manage inventory levels more effectively to reduce the risk of unhealthy inventory.
[0014] Customer Demand Analysis: Customer Demand Analysis involves using customer data to identify buying trends and preferences. By analyzing customer demand data, businesses can identify which products are popular and which are not, and adjust inventory levels accordingly to avoid excess or unhealthy inventory.
[0015] Stock Replenishment Forecasting: Stock Replenishment Forecasting involves using historical data to forecast when stock needs to be replenished. By forecasting stock replenishment needs, businesses can avoid overstocking or understocking and reduce the risk of unhealthy inventory.
[0016] Product Life Cycle Analysis: Product Life Cycle Analysis involves analyzing sales data to identify which products are at what stage in their life cycle, such as the introduction, growth, maturity, or decline phase. By identifying products that are in the decline phase, businesses can avoid overstocking and reduce the risk of unhealthy inventory.
[0017] Seasonal Adjusted Demand Forecasting: Seasonal Adjusted Demand Forecasting involves using historical data to forecast demand, while also adjusting for seasonal factors. By using this method, businesses can better predict demand for products and adjust inventory levels to reduce the risk of unhealthy inventory.
[0018] However, these prediction techniques lack in many at least the following features:
[0019] Lack of Real-time Updates: Many of the prior techniques rely on historical data, which may not reflect the current market dynamics. Unforeseen events or sudden changes in demand can impact inventory health, and relying solely on historical data may not provide an accurate picture.
[0020] Inability to Handle Complexities: Inventory management involves various complexities, such as seasonality, SKU variations, lead times, and pricing dynamics. Some of the prior techniques may not adequately address these complexities, leading to suboptimal inventory decisions and increased risk of unhealthy inventory.
[0021] Lack of Automation: Some of the prior techniques heavily rely on manual analysis and decision-making. This can be time-consuming, prone to errors, and may not scale well for businesses with large or complex inventory portfolios.
[0022] Insufficient Consideration of External Factors: Prior techniques often focus on internal data and metrics, neglecting external factors such as market trends, competitor analysis, or macroeconomic conditions. Failing to account for these external factors can result in inaccurate predictions and ineffective inventory management strategies.
[0023] Lack of Scalability: Some of the prior techniques may not be easily scalable to handle large and complex inventory systems. As businesses grow and expand, managing inventory becomes more challenging, and the limitations of these techniques may become more apparent.
[0024] Lack of Adaptability: Prior techniques may not adapt well to changing business environments or evolving market conditions. They often rely on historical data and assumptions that may not hold true in dynamic and unpredictable markets.
[0025] Limited Insights into Root Causes: While prior techniques can identify unhealthy inventory, they may not provide deep insights into the underlying causes of the problem. This can make it difficult to develop effective strategies to address the root causes and prevent the recurrence of unhealthy inventory in the future.
[0026] Lack of Collaboration and Communication: Some of the prior techniques may not facilitate effective collaboration and communication among different stakeholders involved in inventory management, such as suppliers, sales teams, and warehouse staff. This can lead to information gaps, delays, and inefficiencies in decision-making processes.
[0027] Lack of Real-time Monitoring: Many prior techniques rely on periodic analysis and reporting, which may not provide real-time visibility into inventory health. Real-time monitoring is crucial to promptly identify and address unhealthy inventory situations.
[0028] Neglect of Customer Behavior: Some prior techniques may not adequately consider customer behavior and preferences. Understanding customer demands and trends is crucial for effective inventory management and reducing the risk of unhealthy inventory
[0029] Hence, in view of existing techniques and the problems associated therewith, there is a need to address the problem of end-to-end unhealthy inventory management through a self-learning based intelligent system.
OBJECTIVES OF THE INVENTION

[0030] The primary objective of the present invention is to provide a method and system for end-to-end unhealthy inventory management.
[0031] Another objective of the present invention is to provide a self-learning machine learning model to accurately predict unhealthy products in the inventory stock.
[0032] Still another objective of the present invention is to improve the efficiency in inventory management by monitoring high-risk products and prevention thereof by promoting the sale of such products and avoiding new purchases.
[0033] Other objectives and advantages of the present invention will become apparent from the following description taken in connection with the accompanying drawings, wherein, by way of illustration and example, the aspects of the present invention are disclosed.
SUMMARY OF THE INVENTION

[0034] The present invention relates to a method and system for end-to-end unhealthy inventory management. The method for monitoring inventory health receives a plurality of initial features based on a historical dataset, by a feature selection module configured to process the plurality of initial features into a plurality of selected features. The feature selection module further identifies a plurality of final features by one or more supervised and unsupervised learning models which are configured to process the selected features and provide a supervised list and an unsupervised list. Said supervised and unsupervised models are retrained by a model monitoring and decision engine running in a production environment taking a plurality of historical models, wherein the historical models are provided with a plurality of model parameters and one or more model selection flags to determine performance of the model by generating a score. The generated score is analyzed to evaluate the performance of the model by comparing said score with a predetermined threshold value.
BRIEF DESCRIPTION OF DRAWINGS

[0035] A complete understanding of the present invention may be obtained by reference to the accompanying drawings, when taken in conjunction with the detailed description thereof and in which:
[0036] Figure 1 illustrates a self-learning model with input features and the process.
[0037] Figure 2 illustrates an end-to-end training process of the self-learning model.
[0038] Figure 3 illustrates a Feature Selection Module.
[0039] Figure 4 illustrates a model monitoring and decision engine module.

DETAILED DESCRIPTION OF THE INVENTION

[0040] The following description describes various features and functions of the disclosed methods and systems with reference to the accompanying figures. In the figures, similar symbols identify similar components, unless context dictates otherwise. The illustrative aspects described herein are not meant to be limiting. It may be readily understood that certain aspects of the disclosed methods and systems can be arranged and combined in a wide variety of different configurations, all of which are contemplated herein.
[0041] Accordingly, those of ordinary skill in the art will recognize that various changes and modifications of the embodiments described herein can be made without departing from the scope of the invention. In addition, descriptions of well-known functions and constructions are omitted for clarity and conciseness.
[0042] Features that are described and/or illustrated with respect to one embodiment may be used in the same way or in a similar way in one or more other embodiments and/or in combination with or instead of the features of the other embodiments.
[0043] The terms and words used in the following description and claims are not limited to the bibliographical meanings, but, are merely used to enable a clear and consistent understanding of the invention. Accordingly, it should be apparent to those skilled in the art that the following description of exemplary embodiments of the present invention are provided for illustration purposes only and not for the purpose of limiting the invention.
[0044] It is to be understood that the singular forms “a,” “an,” and “the” include plural referents unless the context clearly dictates otherwise.
[0045] It should be emphasized that the term “comprises/comprising” when used in this specification is taken to specify the presence of stated features, integers, steps or components but does not preclude the presence or addition of one or more other features, integers, steps, components or groups thereof. The equations used in the specification are only for computation purposes.
[0046] The term model or module used herein denotes software, firmware, or hardware components or a combination thereof, performing a specific function of preventing and predicting unhealthy stocks or products in an inventory or warehouse. The model or module may be configured to operate in conjunction with a generic or a specific processing unit to execute instructions to carry out the functioning of the present invention. Other hardware or software components may be utilized to implement the present invention as per the requirements of the user. The term unhealthy in the context of the disclosure of the present invention may be understood in terms of inventory including but not limited to goods, materials, or products, finished or unfinished, capable of being sold by manufacturing companies, or sellers, either online or through conventional means, which are no longer in demand or have exceeded their shelf life, thereby incurring losses to the companies or sellers.
[0047] While this invention has been described in connection with what is presently considered to be the most practical and preferred embodiment, it is to be understood that the invention is not limited to the disclosed embodiments, but, on the contrary, is intended to cover various modifications and equivalent arrangements included within the scope of the appended claims.
[0048] Accordingly, an End-to-End system for unhealthy inventory management may be provided for managing excess products or avoiding stock storage in a warehouse. The term unhealthy inventory refers to products that are no longer in demand or have exceeded their shelf life. Accurate prediction of such unhealthy products or stock keeps an inventory healthy boosts the revenue and also reduces overhead holding costs in maintaining unsold products. Unsold or excess stock may result in several negative consequences for businesses, including stuck working capital, storage occupied by low or slow-selling products leading to high amount of holding cost and inability to keep fast-moving products due to occupied space leading to sales loss, logistic cost and liquidation cost. When this unhealthy inventory is not managed efficiently, the ratio of unhealthy inventory to total inventory may even go up to 40%.
[0049] The cause of unhealthy inventory may be multifold such as overstocking, incorrect forecasting, lack of sales, seasonal demand, product obsolescence, etc. Companies or e-commerce sellers may overstock products in anticipation of high demand or take advantage of bulk discounts from suppliers. However, the excess inventory may become unhealthy if demand does not meet expectations. Also, incorrect forecasting may lead to inaccurate demand forecasting, which may end up with too much or too little inventory, which may result in unhealthy inventory. Incorrect forecasting may occur due to factors such as changes in market trends, customer behavior, or supplier performance. Whereas lack of sales of products also leads to unhealthy inventory. This may occur due to factors such as changing customer preferences, increased competition, or poor marketing. Some products may have high demand during specific seasons or events, such as holidays or sporting events. If companies or e-commerce sellers do not adjust their inventory levels accordingly, they may end up with excess inventory that becomes unhealthy. Product obsolescence is also a major reason for the lack of sales including those products that become outdated or are replaced by newer versions may quickly become unhealthy inventory along with fad products. This may occur due to various factors such as technological advancements, changes in customer preferences, industry trends, etc.
[0050] In accordance with Figs. 1-4, the present invention provides an end-to-end system including a machine-learning based unhealthy inventory model. Fig. 1 illustrates a self-learning model for unhealthy inventory management which processes a product’s dataset to determine the healthy or unhealthy status of the product in the inventory. The model comprises a feature engineering module to receive product information such as data related to sales, price, inventory levels, seasonality, product attributes, etc., and features are rendered based on said data, wherein data profiling is employed to preprocess and clean said features. Further, Exploratory Data Analysis (EDA) is implemented to discover patterns of the rendered features based on summary statistics and univariate/ bivariate analysis. The performance of the trained model is determined based on the level of a score of the model and a threshold value. Based on the comparison, if the performance of the model is below the desired threshold, the model is retrained, and if the performance is above the desired threshold, the results of the model are integrated with the buying system and intervention planner to perform a specific intervention for accurately determining the health status of the product.
[0051] The present invention constructs and extracts one or more features from a relevant dataset corresponding to a product in an inventory. The relevant dataset may be a historical dataset providing details about a product and data related thereto such as, but not limited to, sales data, page views of the product on the website, price, Days on Hand (DoH) data indicating the product’s holding period in an inventory, Disaster Risk Reduction (DRR) data (forward and backwards), days sold, seasonality of the product, age of the product, event days (previous and current month), user ratings or reviews, return date, quality of product, etc. Based on the presented dataset, one or more features are identified and multiple features are created by combining variables together using different variations like percentile groups, rate of change over months, ratios etc. In an exemplary embodiment, the multiple feature selection techniques include a filter method, an embedded method, or a wrapper method. As there are multiple methods and no fixed rule is available to determine the best feature selection method, the present invention provides an innovative approach by selectively combining the methods to determine the best-suited method for a given dataset.
[0052] In accordance with Fig. 2, the unhealthy inventory management model includes a feature engineering module comprising a feature selection module (FSM) and a model training and monitoring system comprising a model monitoring and decision engine module (M2DE). In some embodiments, the a plurality of initial features based on a relevant historical dataset of a product are constructed and extracted by feature construction and feature extraction modules of the FSM . Based on multiple feature selection techniques, the FSM is configured to either select or drop features of the initial features and the selected features are provided to the model training and monitoring system. The model training and monitoring system is configured to process the selected features, and based on an output score generated by the M2DE, it may be determined whether to continue using the model, or retrain the model by the model training and monitoring system, or redevelop the model by the feature engineering module, if the output score of the model is above, below, or in between desired thresholds.
[0053] In accordance with Fig. 3, the feature construction and feature extraction modules respectively construct and extract the initial features from the historical dataset, and based on one or more multiple feature selection techniques, a list of selected features based on the threshold of each such method utilized by the FSM is created. The selected features are provided to the model training and monitoring system, for example, Fig. 2 shows the feature universe with selected features ranging from Feature 1, Feature 2, Feature 3, Feature, 4, upto Feature N, where N is an integer. The feature selection techniques may include, but are not limited to, filter methods, embedded methods, wrappers methods.
[0054] The filter methods are statistical measures which select a subset of relevant features for use in model construction. The filter methods help to avoid irrelevant and redundant features from the model by using different metrics through ranking and therefore implemented for the basic filtering of features. The filter methods, which need low computational time and do not overfit the data, may include methods including but not limited to information gain method, missing value ratio, chi-square test, fisher’s score measures, etc. The information gain method may be used as a feature selection technique by calculating the information gain of each variable with respect to the target variable, and the threshold is less or equal to 0.05. The missing value ratio may be used for evaluating the feature set against the threshold value, and the threshold is less or equal to 6%. The Chi-square test determines the relationship between the categorical variables which is calculated between each feature and the target variable p is less or equal to 0.05. The Fisher’s score is provided for feature selection which computes a score for each continuous feature and ranks them according to their scores. The model considers features with a higher Fisher's Score which is more important. Accordingly, the top 80 features may be filtered and selected.
[0055] The embedded method is provided to learn which features best contribute to the accuracy of a model while the model is created, e.g., regularization methods, random forest importance method, etc. The wrapper method selects a set of features as a search problem and different combinations of sets are prepared, evaluated and compared with other combinations, e.g., forward, backwards, and recursive search methods. Accordingly, the embedded and wrapper methods are implemented and used with the top features to be selected based on the top N features. Once a laundry list of features is obtained by running multiple feature selection methods, the present invention further trains multiple linear and non-linear supervised and unsupervised models to arrive at the top contributing features so that all the features may contribute to decision-making.
[0056] The supervised and unsupervised models may be implemented to generate a final features list from the initial features. Said models are configured to train on the selected features of the historical and latest datasets to capture current trends so that a supervised and an unsupervised feature lists may be generated. In some embodiments, the supervised model may be a linear model or a nonlinear model, whereas the unsupervised model is a variable clustering based model or an auto labelling model. In an exemplary embodiment, the supervised model include a CatBoost predictive model, RF-based predictive model, and gradient-boosting based XgBoost predictive model.
[0057] In accordance with Fig. 4, the present invention provides the model training and monitoring system comprising a model monitoring and decision engine (M2DE) to evaluate performance of new models. In some embodiments, the M2DE is configured to retrain the supervised and unsupervised models on a plurality of historical models, wherein the historical models are provided with a plurality of model parameters and one or more model selection flags to determine an optimum new model configured to generate a score, wherein the generated score is utilized to identify a result based on a predetermined threshold. The result determines a subsequent action for the new model and the action includes continuation, retraining, and redevelopment of said model. The M2DE is configured to run in a production environment wherein a plurality of new data is provided, and the historical models running in the production environment may be continued to use, or retrained, or redeveloped from the scratch, based on one or more corresponding predetermined thresholds.
[0058] The feature selection is done using the feature selection methods and the list of final features is further provided as input into two separate models i.e., the supervised (ensemble) and unsupervised (auto-labeling framework). While maintaining a historical repo of key performance parameters of the models such as F1-Score, Precision, Recall, etc., as shown in Fig. 4, running in the historical timeframe, thresholds for each of the parameters based on the historical patterns are determined. Said thresholds may be known as desired historical or predetermined thresholds having upper and lower thresholds. The upper and lower thresholds may vary for each key performance parameter based on specific use case. The current output or score of the model is compared against historical threshold values to decide on the final prediction. For example, if a score is equal to or greater than an upper threshold, the latest model is allowed to be used continuously. Whereas, if the score is lower than the upper threshold and greater than the lower threshold, then the model is retrained using the same features with different sample sets. If the score is lower than the lower threshold, then the feature selection framework is reviewed as the features are not adding any value. Accordingly, once a final model is chosen, said model is pushed into production and model results are saved in a historical table.
[0059] The present invention provides a technical solution to the problem of end-to-end unhealthy inventory management through an intelligent system. The present invention provides the technical solution based on a self-learning machine learning model at its core that predicts products or stocks which are likely to become unhealthy and determines products having the highest probability of becoming unhealthy in the near future. This helps to improve the efficiency in inventory management which enables managers to identify high -risk products to take appropriate actions accordingly. Early unhealthy prediction of unsold or surplus products helps in avoiding fresh purchases and enables managers or sellers to take proper actions to prevent losses, such as providing additional discounts to the customers for faster sales and timely disposal of unhealthy products before these products stop selling or become redundant. Sellers may liquidate these products to external agencies or return these to the seller from whom they purchased the stock. In effect, the present invention helps to decide which products should not be bought more or which products should manufacturing companies or sellers get rid of quickly so that they don’t become a burden causing losses to them. This will help to proactively save the working capital of companies or e-commerce sellers from getting stuck in non-selling products, help them avoid space occupancy, and improve the profitability or bottom line of the business.
[0060] The present invention provides the advantages of high accuracy, self-learning and providing automated mechanisms for dealing with scalable and complex inventory systems. The disclosed system of the present invention is powered by a scientific machine learning model with very high accuracy as compared to any other technique or manual processes. In some embodiments, the system may work on real-time data, wherein the system executes desired operations independently and also considers complex variables to make decisions. The system is self-learning, i.e. it is able to account for changes in the model accuracy. If the system performance is not above a certain threshold, the system re-trains itself on the inputs available and also finds out which Machine learning technique is most suitable. Moreover, it handles scalable and complex inventory systems automatically. It is easily scalable and addresses challenges that come with large and complex inventory systems where there are many complexities, such as seasonality, SKU variations, lead times, and pricing dynamics.
[0061] The embodiments disclosed herein include systems and methods which may be performed through multiple operations or steps by hardware, software, firmware and/ or any combination thereof. The embodiments may be embodied in machine-executable instructions stored in a memory which may cause one or more general-purpose or special-purpose processors to perform the desired operations of the present invention. Processor such as CPU, GPU, or any other processing unit that may be implemented by different types of electronic components such as logic circuits, microprocessors, Integrated Circuits, microcontrollers, etc., may be used in the present invention as per the requirement. Software may include, but is not limited to, a programming interface that enables different software components to communicate with each other, whereby internet-based web or mobile applications may access or request remote web services, e.g., through their application programming interfaces.
[0062] In an embodiment, the present invention may include an end-to-end system for monitoring inventory health by a machine learning based unhealthy inventory model, comprising a memory configured to store a plurality of machine-executable instructions of the model and at least a processor configured to execute said instructions to perform operations and thereby implementing the steps of the model.
[0063] In an embodiment, the present invention may include a method for monitoring inventory health by an end-to-end machine learning based unhealthy inventory model, comprising the steps of: receiving, a plurality of initial features based on a historical dataset, by a feature selection module to process the plurality of initial features into a plurality of selected features; identifying, a plurality of final features, by one or more supervised and unsupervised learning models by processing the selected features and providing a supervised list and an unsupervised list; and retraining said supervised and unsupervised models by a model monitoring and decision engine running in a production environment on a plurality of historical models, wherein the historical models are provided with a plurality of model parameters and one or more model selection flags to determine performance of the model by generating a score, wherein the generated score is analyzed to evaluate the performance of the model by comparing said score with a predetermined threshold value.
[0064] In an embodiment, the unhealthy inventory model may include a feature selection module configured to select one or more features based on one or more feature selection techniques and a corresponding threshold value assigned to each said feature selection technique, wherein the feature selection technique is selected from filter method technique, embedded method technique, wrapper method technique.
[0065] In an embodiment, the unhealthy inventory model may include a supervised model which is selected from a list of linear models and non-linear models, and the unsupervised model is selected from a list of variable clustering model, auto labelling model.
[0066] In an embodiment, the unhealthy inventory model may include predetermined threshold value which is a historical threshold value of key performance parameters including F1-score, Precision, and Recall, of one or more models running in the historical timeframe for each corresponding parameter based on its historical patterns.
[0067] In an embodiment, the unhealthy inventory model may include a score which determines a subsequent action for the model and the action includes continuation, retraining, and redevelopment of said model based on the predetermined threshold value, the predetermined threshold value includes upper and lower threshold values, wherein the score is equal to or greater than the upper threshold value and the model is in continuation as the current model in use, the score is less than the upper threshold value and greater than a lower threshold value and the model is further retrained on the same features with different sample sets, and the score is less than the lower threshold value and the model is a model where the features are redeveloped by the feature selection framework.
[0068] In an embodiment, the unhealthy inventory model may include a historical dataset which includes information about a product in inventory and comprises a plurality of data including but not limited to sales data, page views on the website, price, Days on Hand (DoH) data indicating the product’s holding period in an inventory, Disaster Risk Reduction (DRR) data (forward and backwards), days sold, seasonality of the product, age of the product, event days (previous and current month), user ratings or reviews, return date, quality of product.
[0069] In an embodiment, the unhealthy inventory model may include supervised models which include a CatBoost predictive model, RF-based predictive model, and gradient-boosting based XgBoost predictive model.
[0070] In an embodiment, the present invention may include a computer program product that may comprise computer-readable instructions for causing a processor to carry out aspects of the present invention on a computer or a processing device.
[0071] While this invention has been described in connection with what is presently considered to be the most practical and preferred embodiment, it is to be understood that the invention is not limited to the disclosed embodiments, but, on the contrary, is intended to cover various modifications and equivalent arrangements included within the scope of the appended claims.
, Claims:WE CLAIM:

1. A method for monitoring inventory health by an end-to-end machine learning based unhealthy inventory model, comprising the steps of:
receiving, a plurality of initial features based on a historical dataset, by a feature selection module to process the plurality of initial features into a plurality of selected features;
identifying, a plurality of final features, by one or more supervised and unsupervised learning models by processing the selected features and providing a supervised list and an unsupervised list; and
retraining said supervised and unsupervised models by a model monitoring and decision engine running in a production environment on a plurality of historical models, wherein the historical models are provided with a plurality of model parameters and one or more model selection flags to determine performance of the model by generating a score, wherein the generated score is analyzed to evaluate the performance of the model by comparing said score with a predetermined threshold value.

2. The method as claimed in claim 1, wherein the feature selection module is configured to select one or more features based on one or more feature selection techniques and a corresponding threshold value assigned to each said feature selection technique, wherein the feature selection technique is selected from filter method technique, embedded method technique, wrapper method technique.

3. The method as claimed in claim 1, wherein the supervised model is selected from a list of linear models and non-linear models, and the unsupervised model is selected from a list of variable clustering model, auto labelling model.

4. The method as claimed in claim 1, wherein the predetermined threshold value is a historical threshold value of key performance parameters including F1-score, Precision, and Recall, of one or more models running in the historical timeframe for each corresponding parameter based on its historical patterns.

5. The method as claimed in claim 1, wherein the score determines a subsequent action for the model and the action includes continuation, retraining, and redevelopment of said model based on the predetermined threshold value, the predetermined threshold value includes upper and lower threshold values, wherein
the score is equal to or greater than the upper threshold value and the model is in continuation as the current model in use,
the score is less than the upper threshold value and greater than the lower threshold value and the model is further retrained on the same features with different sample sets, and
the score is less than the lower threshold value and the model is a model where the features are redeveloped by the feature selection framework.

6. The method as claimed in claim 1, wherein the historical dataset includes information about a product in inventory and comprises a plurality of data including but not limited to sales data, page views on the website, price, Days on Hand (DoH) data indicating the product’s holding period in an inventory, Disaster Risk Reduction (DRR) data (forward and backwards), days sold, seasonality of the product, age of the product, event days (previous and current month), user ratings or reviews, return date, quality of product.

7. The method as claimed in claim 1, wherein the supervised models include a CatBoost predictive model, RF-based predictive model, and gradient-boosting based XgBoost predictive model.

8. An end-to-end system for monitoring inventory health by a machine learning based unhealthy inventory model as claimed in claim 1, comprising a memory configured to store a plurality of machine-executable instructions of the model and at least a processor configured to execute said instructions to perform operations and thereby implementing the steps of the model.