Overview / Usage

A Just in time (JIT) production is an idealized concept where we supply the required material at the desired place at the required time. Just-in-time (JIT) production methods have been proven to be an efficient way of manufacturing. It reduces the various costs involved in a supply chain, such as cost of warehousing, security and insurance, increasing throughput and yield. Restaurants and other food delivery businesses are perhaps one of the prime examples of just-in-time production environment, wherein the food items are produced only after an order is received and delivered at the required place in very little time.

Just-in-time (JIT) is an idealized concept is Supply Chain Management (SCM) where no inventory needs to be managed and required inventory is supplied at the place at the required time. This has may benefits which include but are not limited to reduction in size of inventory, reduction in warehousing or storage cost, increased throughput and more efficiency. This system however places some requirements on the vendor, such as having sufficient capacity to supply anytime without passing cost of overcapacity to buyer. It also requires that the vendor takes into account the total lead time for the deliveries, which in itself is a function of various times, such as processing time, breakdown time, fixing times among others. Machine Learning and Artificial Intelligence can be applied to almost every field of Supply Chain, and has the capacity to drastically improve the supply chain, and help in creating new models of the Supply Chain which take into account uncertainty, and hence can be applied to real life, non deterministic environment. J. Mula et al. reviewed the various applications of Artificial Intelligence in Supply Chain Management, such as in aggregate planning, material requirement planning, manufacturing resource planning, inventory management and supply chain planning. In this paper, we study a particular case of the JIT system, which are the restaurants. Restaurants supply the required order to the customer when and where needed by them. Delivering order to their customers on time is perhaps one of the foremost objectives of restaurants. In order to achieve this, they must take into account the order lead times too, and should be able to predict it accurately. They must also identify the features on which this order lead time depends, and use them to train a model for predicting it accurately.

Methodology / Approach

Before fitting the data into a Machine Learning model, we need to gather more information about the data. This includes the information related to the description of raw data, and also the statistical information about the data. We need to fill null or empty values with suitable values such as mean or median of the data. Visualization of the data needs to be done if required, by making different scatter plots, histograms etc, to see a pattern in the data. To avoid curse of dimensionality, we select only features that have high correlation with the target variable. To find the correlation between the target value, i.e., the order lead time and features by comparing values of Pearson, Kendall and Spearman coefficient of correlation. We select the features that have highest correlation to target variable and fit them to various Machine Learning model. Then we use this model to make predictions on test set, calculate various errors on the results obtained and compare the results obtained by various models.

Correlation measures how much a specific attribute depends on other attribute. Often we use Pearson’s coefficient of correlation to measure the linear correlation between different attributes. The value of correlation coefficient varies from -1 to 1. A value of 1 indicates a perfect positive correlation, i.e., as one attribute increases, the second attribute increases. Conversely, a value of -1 indicates a perfect negative correlation, i.e., as value of one attribute increases, the value of other attribute decreases. A value of 0 indicates no linear correlation between the two attributes. However, there may still be a non- linear correlation, which can be measured with the help of Kendall’s coefficient of correlation and Spearson’s coefficient of correlation.

We fit the data to various machine learning algorithms, and the predictions are compared with true value, and errors are measured with various techniques. The various techniques used for measuring errors are Mean Absolute Error (MAE), Mean Squared Error (MSE), Root Mean Squared Error (RMSE) and Mean Absolute Percentage Error (MAPE). Values of various parameters such as selection of value of k in k-NN and selection of maximum no. of leaves in decision tree were chosen by selecting the most optimal value of the parameter that minimizes the errors. The following results were obtained after evaluating different models on the test set and measuring errors from different techniques.

Technologies Used

Python

Numpy

Pandas

Matplotlib

Scipy

Scikit-Learn

Jupyter Notebook

Other links

• Project Link