Help regarding an ETA prediction problem

High Loss in Neural Network Training for Delivery Time Prediction.
Hey guys!! ,
I am trying to build a system which is capable of predicting ETA for a delivery organisation . I have historical data which has informations regarding the client who ordered , the delivery guy’s name , the time the order was recived at , the time the order was shipped at , and the order at which the order was recived , the location inofrmation of the shop and the location information of the client .

The labels are the following order

id client_name shop_name shop_location delivery_location order_received order_shipped_at order_delivered_at delivery_boy

the following is an example row:-

client name shop name 26.09149940481357, 43.98838843280797 26.1108475,43.9617274 16-05-2023 12:47 16-05-2023 13:16 16-05-2023 13:17 JHON biker

Now i had uniquely encoded the names and string charecters and seperated the longitudes and latitiudes and convereted the time into seconds and then tried to give all of these as the input into a nerual network which should have given a single output in second which would be the time of arrival in seconds. I used a simple neural network inorder to d this but it isnt working and im getting a very high error value .

Please help me out in this issue, and please do ask for any clarifications that would be needed in this aspect.

First suggestion is to use elapsed time, instead of absolute time.

Can you share some details that how you are tranning and measuring the error.